从湿实验室到强化学习环境：一个周末完成一万次药物发现——Ian Quigley，Leash Bio

我和我的联合创始人 Andrew Blevens 在 Recursion 时，正看着屏幕上的一个小分子，然后一位药物化学家走过，名叫 Mike Karesi，Mike 说，嘿，看起来像 Stuart Tribert 的分子。然后我们想，哦不，你光走过去就能看出这是谁做的？太可怕了。如果模型仅通过观察分子就能识别化学家，那么也许模型能推断化学家的意图，这意味着也许它们什么都没学到。

My co-founder Andrew Blevens and I were at Recursion looking at a small molecule on a screen, and then a medicinal chemist walked by us, this name Mike Karesi, and Mike was like, hey, looks like a Stuart Tribert molecule. And then we were like, oh no, you can tell just by walking by and be like who made this? That's horrifying. And if the models can identify the chemist by looking at the molecule alone, then maybe the models can infer chemist intent, which means maybe they don't learn anything at all.

嗯，它们确实学到了一些东西，是的，它们不学物理。

Well, they did learn something, yeah, they don't learn physics.

我们从 Kemble 下载了所有化学家的所有化学品，这是这些及其活性的最大存储库。每个发表超过，我想是 30 篇论文和 600 个分子的化学家。所以你得到了所有分子和所有东西。有 1,815 位化学家。然后我们训练了一个分类器，仅通过观察分子来预测化学家。我们得到了前五准确率 60%。我们选择不用任何化学结构或公式来表示分子，而只是一个创造者概率的向量。所以现在它是一个 1,800 维的向量，有某个人制造它的概率。你可以使用这个表示，然后训练一个模型来预测 Kemble 的活性，它的表现和使用化学公式一样好。我们真的认为它学到了很多社会学。它没有学到 ground truth。当你在尾气上训练模型时，你可能会遇到麻烦。

We downloaded all chemicals from all chemists at Kemble, which is the biggest repository of these things and their activities. Every chemist who published more than, I think it was 30 papers and 600 molecules. So you get all the molecules and all the stuff. It's 1,815 chemists. And then we trained a classifier to predict a chemist from looking at the molecule alone. And we got a top five accuracy of 60%. We chose to represent the molecules not with any chemical structures or formulas at all, but just a vector of the probability of creator. So now it's this 1,800 element vector and it's got some probability of who made it. And you can use that representation and then train a model to predict activity from Kemble and it does as well as if you use chemical formulas. We really think that it's learning a lot of sociology. It's not learning ground truth. When you train models on the exhaust, you might be in trouble.

欢迎来到 Laten Space AI for Science 播客。我是 R.J. Haniki，Mirror OMIX 的首席技术官，和我一起的是我的联合主持人 Brandon Anderson。他在 Ato atomic AI 开发 RNA 疗法。

Welcome to Laten Space AI for Science podcast. I'm R.J. Haniki, CTO of Mirror OMIX and I'm joined by my co-host Brandon Anderson. He develops RNA therapeutics at Ato atomic AI.

非常兴奋今天能请到 Ian Quigley 来到演播室。今天，这一集将会很精彩。非常激动。嗯，Ian 是 Leash Bio 的联合创始人兼 CEO。呃，在我看来，这是过去几年我最喜欢的年轻科技生物公司之一。你知道，在高层次上，嗯，Leash 结合了认知严谨性、疯狂规模的数据收集，并且，呃，对开放科学有坚定的承诺，还有一系列令人难以置信的博客文章，我们今天会涵盖很多，但我认为这些文章真正告诉你如何在非常困难的领域以正确的方式做机器学习。我想我会有些有趣的轶事，希望能在这一集中穿插。但是，是的，首先，呃，感谢你来到这里，呃，你知道，你的，呃，在我们开始之前，你有什么关于我们想告诉我们的吗？

Super excited to have Ian Quigley in the studio with today. Today, this is a this episode's going to be a treat. Very excited. Um Ian is the co-founder and CEO of Leash Bio. Uh which in my personal opinion is one of my favorite young tech bios of the last like few years. you know, at a high level, um, Leaf is combining epistemic rigor, data collection at like a crazy scale and, uh, has has a strong commitment to open science and has an incredible series of blog post, a lot of which we'll cover today, but which I think really tells you how to do machine learning in a very hard domain in kind of the right way. I think I'll have some fun anecdotes hopefully can uh, trickle in through the episode. But yeah to start with uh thanks for being here and uh you know what's your uh is there anything about us that you want to tell us before we get started?

嗯，不，谢谢，你太客气了。是的，我们，呃，我们创立 Leash 的历史上，嗯，是在遇到一个特定问题之后，即试图预测小分子与蛋白质靶点的结合，呃，然后发现现有的数据有些不足，所以我们出去以合理的规模构建了这个东西，试图去攻克它。

Um no thanks that was very kind. Yeah we uh we founded Leash uh historically um after being confronted by a particular problem of trying to predict small molecules binding to protein targets uh and then finding that the data that was out there was kind of wanting and so we went out and built this thing at a reasonable scale to kind of try to go after it.

是的。

Yeah.

是的。所以合理的规模。嗯，你们的创立基于做一种叫做 DNA 编码库的东西。我见过的每一个接触过 DEL 的人出来都严重受创。我不认识任何人说过 DEL。那是个好主意。让我们做数百或数千个这样的东西。所以为什么？而且在你开始之前，请解释一下什么是 DEL。

Yeah. So reasonable scale. Um, you're founded upon the idea of doing something called a DNA encoded library. Every person I've ever met who has touched a Dell has come out severely traumatized. I I don't know any person who has ever said Dells. That's a great idea. Let's do hundreds or thousands of these things. So why what? And also please explain what a Dell is before you get started.

是的。是的。呃，所以当你进行小分子药物发现时，呃，最终目标是让一个小分子或微小的化学物质结合到你认为在疾病中引起问题的蛋白质上。呃，然后从那里开始。呃，其中一些蛋白质很容易找到粘附的化学物质，有些则更难。一种方法是把许多不同的化学物质放在不同的管子里。呃，然后你一次将蛋白质浸入一个管子中，这几乎要花很长时间。呃，然后几年前提出的另一种解决方案是，也许如果你把很多化学物质放在一个管子里，然后每个小分子都物理连接一个小 DNA 条形码。你像鱼钩一样将蛋白质浸入管中。你拉出结合的不同分子，呃，然后对这些分子的 DNA 进行测序。所以这个想法是，你可以拉出大量分子，然后你就不必承担跨越大量试管或单个孔和板的物流开销。所以下游发生的事情是，你有化学库，如果一次一个管子，世界上最大的在低个位数百万。使用 DEL，已经产生的最大的是 40 万亿个分子。呃，当它们变得那么大时会有一些问题，但这是你可以尝试用这种技术达到的规模。所以你创建这些东西的方式是，你使用组合化学，你有这些构建块，然后你一块一块地构建库，但你是组合地构建它。所以

Yes. Yes. Uh so when you're going after a small molecule drug discovery, uh the ultimate goal is to get a small molecule or a dinky chemical to bind to the protein that you think is causing the problem in a disease. Uh and then go from there. Uh and some of these proteins are pretty easy to find chemical material that sticks to you and some are harder. One way to go about this is to have a lot of different chemicals in different tubes. Uh and you then dunk the protein in one tube at a time and it kind of takes forever. Uh, and then a different solution that was proposed uh, some years ago now was maybe if you had a lot of chemicals in the tube at once and then one tube and then you've got a small DNA barcode that's physically attached to each small molecule. You dunk your protein in the tube kind of like a fish hook. You pull out different ones that bind uh, and then you sequence the DNA of those. And so the idea is that you could then pull this large number of molecules and then you wouldn't have the logistical overhead of trying to go across large numbers of tubes or individual wells and plates. And so what's happened downstream of that is that you've got chemical libraries that if you do one tube at a time, the biggest ones in the world are in the low single-digit millions. With Dell's, the largest one that has been produced um is 40 trillion molecules. Um there's some problems when they get that big, but that is the kind of scale that you could try to attempt with this technology. So the way you create these things is you you uh use this combinatorial chemistry where you have these building blocks and then you you build the library piece by piece but you build it combinatorilially. So

没错。是的。是的。所以稍微解释一下，呃，如果你有数百万个分子，这实际上意味着它是构建块的组合混合物，构建块的数量会比所有最终产物小得多。所以可能是平方或立方，对吧？呃，在一些不同的例子中。所以有数千或数万个构建块，然后生成数百万个分子的化学库并不罕见。这在某些方面是限制性的，因为你知道你被困在较少数量的构建块中。但对机器学习来说，一个非常好的事情是，你有所有这些不同的组合，它们看起来非常相似，除了可能有一两处变化。所以，你知道，我肯定稍后会谈到这个，呃，但这确实使得比较成为可能，这样你可以识别，哦，是小分子的这一部分驱动了结合，因为我看到另一部分有很多变化，但没有太大影响行为。呃，这被证明是非常有用的。

That is correct. Yes. Yes. And so to unpack that a little bit uh if you've got you know some number of millions of molecules what this really means is it is a combinatorial mixture of building blocks and the number of building blocks is going to be a lot smaller than the final product of all of those. So you might be squared or cubed, right? Uh in some different uh example. So it's not uncommon for there to be maybe thousands or tens of thousands of building blocks that would then generate a chemical library of many millions of molecules. And this is restrictive in some ways because you know you're stuck with a smaller number of building blocks. But one thing that's really nice about it for machine learning is that you've got all these different combinations of things that look very similar to each other except for maybe one or two changes. And so that, you know, I'm sure going to get into this later, uh, but that does really enable comparisons so that you can identify, oh, it's this part of the small molecule that's driving the binding because I see a lot of variations in this other part that's not affecting the behavior too much. Uh, and that turns out to be incredibly useful.

另一个方面是你如何进行读出，即 DNA 测序。这在本播客中一直是一个大主题。我们有很多嘉宾，他们都有某种变体，我们做一些疯狂的事情，然后我们将读数转换为 DNA 测序读出。那么这是如何工作的，呃，为什么这么有用？

The other aspect of this is how you do the readout, which is this DNA sequencing. This has been a big theme on the podcast. We've had many uh guests who have had some variation of we do something crazy and then we convert the reading into a DNA sequencing uh readout. So how does this work and uh why is this so useful?

对。是的。我绝对会说，在 Leash 有一条路线，如果你要去做生命科学的东西，呃，并将其与机器学习方法结合起来。真的只有一个可扩展的技术，那就是拍照。测序 DNA 是另一种拍照方式。呃，所以基本上就是拍照和 DNA 测序。

Right. Yeah. I would definitely say that it is a party line at leash that if you're going to go off and try to do uh life science stuff uh and couple it with machine learning approaches. There's really only really only one scalable technology in town and that technology is taking pictures. Sequencing DNA is another way of taking pictures. Uh and so it's basically pictures and DNA sequencing.

如果你能把你的难题塞进那套流程，用其中一种方法作为读出手段。比如 Recursion 直接拍细胞的照片，还有一堆其他团队也这么做，像 INRIA 就这么干；然后几乎其他所有人用空间混合那套或者别的路子，他们会把拍照和测序混着来，而测序其实也是拍照。对，肯定有些蛋白质组学的人会不同意我，但我还是认为规模还没到。

And if you can shovel your challenging problem into that using one of those approaches as a readout. So for you know like Recursion takes actual pictures of cells and a bunch of other groups do that, like INRIA does that, and then almost everyone else with the spatial mix stuff or other approaches, they'll do some mixture of pictures and sequencing, which is also pictures. Yeah, there'll be some proteomics people out there who are going to disagree with me, but I'd still argue the scale is not there yet.

第一个劲爆观点。

Hot take number one.

对。测序最酷的一点就是它便宜得离谱。它极度可扩展，是非常成熟的技术，我们非常擅长它。

Yeah. And one of the really cool things about sequencing is that it is dirt cheap. You can, you know, extremely scalable, very well-established technique, like we're very good at it.

它确实非常非常便宜，而且一直在变便宜。有一小段时间价格下降有点停滞，但现在一堆新公司冒出来，又把价格压下去了。

It is very, very cheap. It is getting cheaper all the time. There was a little while where it kind of plateaued in terms of the price dropping, but now a bunch of new companies have started up and they're driving it down again.

所以我们非常关注测序成本下降的轨迹，这会影响我们如何看待数据采集以及我们做的规模。对，这真的很好。

And so we very much look at the trajectory of sequencing cost dropping, and it has implications for how we think about data collection and the scale that we're doing. Yeah, it's really nice.

这很酷。所以你有这些组件，把它们混在一起形成这种组合式的分子混合物，而每个组件都带条形码。是这样吗？

That's cool. So you have these components that you are mixing together to form this combinatorial mixture of molecules, and each of those components is barcoded. Is that right?

没错。然后它们会彼此结合，因为它们是这些蛋白链或氨基酸链，会咔嗒一下接到彼此的末端并折叠在一起，然后它们要么结合你的靶标，要么不结合。

That's correct. And so then they kind of bind together because they're these protein chains or these amino acid chains that will click onto the ends of each other and fold together, and then they either bind or don't bind to your target.

那么首先，你怎么知道它们结合到了你的靶标上？

So first of all, how do you know that they bound to your target?

对。首先，这非常像鱼钩的类比：现在靶标是某个蛋白或蛋白复合物，你要试着找到能与之结合的化学物质。它被固定在一个固相载体上。有人用各种不同的方式，但我们通常的做法是把它固定在一颗磁珠上。然后你把这颗珠子浸到所有这些小分子组成的复杂混合物里。在最简单的情况下，假设这个蛋白上有一个形状非常独特的洞。你把它浸进去，然后那个池子里的化学物质中，有一小部分具有合适的形状和电子特性，能结合进这个形状独特的洞里。然后你把这个蛋白拉出来，你只拉出那些在洞里的东西。你再用某种方式冲洗几次。然后你对那些出来、在洞里的东西的 DNA 进行测序。

Yeah. So first of all, it is very much like this sort of fish hook analogy, where if you were to — so right now the target is some protein or a protein complex that you're going to try to find chemical material to bind to. It's attached to a solid support. And so some folks do it a variety of different ways, but the way that we usually do it is it's attached to a magnetic bead. And so you dunk this bead in this complex mixture of all these small molecules. And in a very easy case, let's pretend that the protein has a very distinctively shaped hole in it. You dunk it in there, and then the chemicals in that pool, some tiny fraction of them, have the appropriate shape and electronics that they're going to bind into this very distinctly shaped hole. Then you pull this protein back out, and you only pull out the things that are in that hole. You rinse it a couple times in some way. And then you sequence the DNA of those things that come out that are in that hole.

好？然后你说对 DNA 测序，你基本上就是在扩增里面的条形码。

Okay? And then when you say sequence the DNA, you're basically amplifying the barcodes that are in there.

没错。然后你就可以——扩增指的是 PCR 之类的。

That's correct. And so then you can — so amplify meaning PCR or something like that.

你做一点 PCR，然后对，然后把它丢进测序仪。

You do a little bit of PCR and then, yeah, then you toss it into the sequencer.

然后你现在可能在想的一件事是：好，我们需要拿它跟某个东西比较，才能很好地衡量相对富集。所以你确实得设计一堆好的阴性对照，来说：哦，如果我只把珠子单独放进去，或者有没有泛蛋白黏附信号会把某些化学库里的化学物质拉出来？所以你得非常认真地思考所有这些，并跟你真正想找到结合物的那个东西做合理的比较。

And then one thing that you're probably thinking right now is, okay, well, we need to compare it to something to get a good measure of relative enrichment. And so you do have to engineer a bunch of good negative controls to say, oh, well, if I just stuck the bead in there by itself, or are there pan-protein stickiness signals that are going to pull out certain chemical libraries, chemicals from the library? And so you got to think pretty hard about all those things and make a reasonable comparison to the thing you're actually trying to get binders for.

对于那些结合的，你就有这些条形码的组合，它们都——好，如果这个结合了，那我就有这些——一个蛋白我们说的是多少个条形码？

For the things that bind, you sort of have these combinations of barcodes that all sort of — okay, if this one bound, then I have like these — how many barcodes are we talking about for one protein?

好问题。一个小——

Great question. One small —

对一个小分子。

For one small molecule.

有几件事。第一，如果我们想深入 PCR 的那些细节——对在座的听众说，如果你在制造大量 DNA 拷贝，DNA 里会有一些细微的偏好，可能让某些拷贝比其他扩增得更快。所以你要控制这一点。你放一点独特的序列进去。这样当你有很多拷贝时，你可以把它们合并，让它们都来自这一个——你知道，就只来自这一个事件。然后还有一堆关于条形码的其他问题，我们很在意，就是当你试着把蛋白放进去时，有些蛋白会黏到 DNA 上，因为那是它们的工作，对吧？有很多蛋白的工作就是与细胞里的基因组相互作用，把东西打开或关掉，或以其他方式影响它们，然后你有了这个实验，里面有大量的 DNA，你把它们扔进去。

So there's a couple things. One thing, if we want to get into the kind of PCR-based weeds — and so for the audience here, if you're making lots and lots of copies of DNA, there's little subtle biases in the DNA that might make them amplify faster than others. And so you want to control for that. So you put in a little bit of unique sequence in there. So when you have lots of copies, you can then collapse it so that they all came from this one — you know, it's very much just from this one event. And then there's a bunch of other things about barcodes that we worry about at length, which is that when you are trying to stick proteins in there, some of those proteins are going to stick to DNA because that's their job, right? There's lots of proteins whose job it is to interact with the genome in cells and turn things on or turn things off or affect them in some other way, and then you've got this assay where you've got a ton of DNA and you're throwing them in there.

对。

Yeah.

所以那可能有风险。

And so that can be risky.

那会导致假阳性，对吧？

And that can cause false positives, right?

非常是。很多人抱怨这个。这有点像之前说的有人做 Dell 噩梦。这就是那些问题之一。有各种方法解决它。我们做的一件事是，我们有大量条形码，每个都编码一种特定类型的小分子。你说这是一个小分子。很好。现在我们用 32 或 48 个不同的条形码都编码这一个小分子。

Very much so. A lot of people complain about this. This is kind of to the earlier comment about folks having Dell nightmares. This is one of those problems. There's a variety of ways to address it. One thing that we do is we have large numbers of barcodes that each encode one particular type of small molecule. You say here's one small molecule. That's great. Now we have 32 or 48 different barcodes all encoding this one small molecule.

哦，我明白了。

Oh, I see.

对。然后你知道——比如会有单个例子，一个单分子有这个条形码，然后它的邻居，化学式相同，会有不同的条形码，然后以此类推——

Right. And so then you know — so like there'll be a single example, one single molecule has got this barcode, and then its neighbor, which is the same chemical formula, will have a different barcode, and then so on for —

所以如果那里有差异，那你就知道你是结合到了 DNA 上，而不是——

So if there's differential there, then you know that you're binding to the DNA instead of the —

没错。所以对在座的 DNA-蛋白爱好者来说，一个非常好的地方是，因为那里有 DNA，你可以看到是否有某个单一的 DNA 编码条形码结合得比所有其他都好，尽管它们都是同一个小分子，然后你可以看你在那里富集到的序列，你就能还原出那些蛋白已知的 DNA 结合偏好。

That's right. And so for the DNA-protein aficionados in the audience, one really nice thing about this is that because there's DNA in there, you can see if there's a single DNA-encoding barcode that binds way better than everything else, even though it's all the same small molecule, and then you can look at the sequence that you have enriched there, and you can recover the described DNA binding preferences of those proteins.

就像，哦，这是一个转录因子，它结合这一系列核苷酸，这是描述得很清楚的，你知道，大概六到八个，然后你看——你从这个实验里还原出 DNA 序列，就像，哦太好了，我们找到了共有结合基序。

It's like, oh, here's a transcription factor, it binds this series of nucleotides that is very well described, you know, it's like six or eight of them, and then you look at the — you recover the DNA sequences back from this experiment, it's like, oh great, we found the consensus binding motif.

对。

Right.

所以我觉得我们的蛋白大概做对了。

So I think our protein probably got made right.

你有了这个你识别出的条形码，而且你用刚才说的技术差不多去掉了假阳性。

You have this barcode that you've identified and you've kind of gotten rid of the false positives using the technique that you just said.

所以现在我知道对于某个给定的组合。

And so now I know that for a given combination.

那么对于我所在的一个给定小分子，它结合到我的靶标上，但是你不是有所有这些 DNA 标签、条形码标签挂在你分子上吗？这不会影响……

So for a given small molecule that I'm in, I bind to my target and um and it's but don't you have all these DNA tags, barcode tags like hanging off your molecule? Doesn't that impact the...

是的，它们很笨重。它们很笨重。这是人们对 DELs 提出的另一个抱怨。呃，这里有几个不同的答案，我想说。但非常宏观的哲学答案是，呃，你可能想要运行这类实验的原因有多个。一个原因是，你试图获得某种行为方式的小分子材料，以便推进到药物项目中。所以如果这是你的主要目标，那么你要确保在带有 DNA 条形码和去掉它之间有一定的保真度。这是真的。当我们尝试追求有趣的化学材料时，至少你必须经历在 DNA 上制造化学物质的工作，你必须在这些功能测定中运行它们，有一定程度的损耗，呃，对于在我们手中容易的东西，大约 90% 有效，对于难的东西大约是 50%，所以有一些损失，其他人有不同的经验，但我也会说，如果你考虑为机器学习训练目的这样做，所有这些带有 DNA 条形码的化学物质都在管中，你运行了实验，有些结合了，有些没有，这是你从世界中物理测量的真实信号。然后你只需根据这个条形码的存在来条件化你的模型。呃，然后你继续你的模型训练。所以在我们手中，当我们重新运行实验时，制造新鲜的蛋白质批次，在数月时间内制造新一批化学库，呃，我们大部分情况下得到相同的命中，以相同的顺序。所以它极其可重复和稳健。所以就地面信号而言，我们感觉相当不错。

Yes, they are hefty. They are hefty. This is another complaint that people bring up about DELs. Uh there's a couple different answers here, I would say. But like the the very like 10,000 foot philosophical answer is that uh there are two different or there are multiple different reasons why you might want to run one of these experiments. One reason is that you're trying to get uh small molecule material that behaves a certain way that you're going to progress into drug programs. And so if that is your main goal, then you want to make sure that you're getting some kind of fidelity between having the DNA barcode on there and then having it off there. And that's true. And when we try to pursue interesting chemical material at least like you do have to go through the work of manufacturing the chemical off DNA you've got to run them in these functional assays there's a certain level of attrition uh for things that are easy in our hands about 90% of them work for things that are hard is about 50 so there's some loss other people have different experiences but I would also say that if you're thinking about doing this for machine learning training purposes all these chemicals with their DNA barcodes are in the tube you ran the experiment some of them bound and some of them didn't and that is real signal that you were physically be measuring from the world. And so then you just condition your models on the existence of this barcode. Uh and then you just keep moving on with your model training. And so in our hands when we rerun experiments, so make fresh protein batches, make a new batch of the chemical library over many many months of time, um we get back the same hits in the same order for the most part. And so it's extremely reproducible and robust. And so in terms of like ground signal, we feel quite good about it.

好的。

Okay.

你可能会在那里用假阴性阻塞口袋或其他东西。但你没有，因为你知道 DNA 在那里。它没有结合。你思考的方式是，哦，我在想分子在没有 DNA 时的行为，但我们知道它在那里，我们不会假装它不在。

You could be blocking the pocket or something with your false negative there. But you don't because you know that the DNA is there. It did not bind. The way that you're thinking about it is like, oh, I'm thinking about how the molecule will behave missing the DNA, but we know it's there and we don't pretend that it's not.

哦，我明白了。那么，你的命中包括 DNA 吗？

Oh, I see. So, does your hit include the DNA then?

是的。就像我们训练模型时，我们非常多地放入一个占位符，呃，标记 DNA 的位置。

Yeah. Like when we train models, we we put we very much put in a placeholder uh that that marks like the position of the DNA.

如果你要为药物合成这个，你会留下那个吗？

If you're going to synthesize this for a drug, will you leave that?

不。就像我说的，那就是你得到那些损耗率的地方，对吧？然后你必须制造它，有时它们行为不同。不总是，但有一部分时间。

No. Like I said, that's that's where you get those attrition rates, right? Then you've got to make it and and sometimes they behave differently. Not always, but some fraction of the time.

明白了。明白了。明白了。

Got it. Got it. Got it.

所以，只需要理解这一点。

So, just need to understand that.

实际上，那是——那些是非常好的问题。

Actually, that was those are really good questions.

那些真的很好。是的。就像有很多反对意见，而且有一个重新——这些是许多人试图做我们正在尝试做的事情的许多原因。

Those are really good. Yeah. Like there there are lots of objections and and like there's one re there. These are many of the reasons why folks are trying to do the thing that we're trying to do.

是的。我正要说回到我最初的说法，我再次从未遇到过一个从事 DELs 工作的人说我们应该做更多这些。

Yeah. I was going to say going back to my original statement, I I again have never met someone who has worked on DELs and said we should do more of these.

所以，是的。是的。

So, Yeah. Yeah.

所以，我——我的意思是，我希望我们会谈到——我想我实际上很兴奋这条路线似乎正在起作用。呃，你知道，希望观众在这结束之前会同意我的看法。事实上，也许甚至一个个人轶事。呃，我们有一个共同的朋友，我想他是 Leash 的天使投资人。当他告诉我他是一家成人公司的天使投资人时，我开玩笑地说：“好吧，那是你再也见不到的钱。” 我在这里，在我们所有观众面前说，我错了。我希望我曾——呃

So, I I mean I I I I think I mean hopefully we will get to I think why I'm actually excited that this route seems to be working. Um you know, hopefully the audience will will agree with me before this is over. In fact, actually maybe even a personal anecdote. Um we have a mutual friend who is I think an angel investor in Leash. And when he told me he was an angel investor in an adult company, I I jokingly said, "Okay, well that's money you'll never see again." And I'm like here in front of our entire audience to say I was I'm I was wrong. I I wish I had uh

我希望我听了。

I wish I had listened.

呃，我首先好奇的是，为什么 DELs 这么难？我想我们刚刚涵盖了很多这些。比如在历史上从事 DELs 的人群中，是什么让它作为一种经历如此创伤性。然后你为什么认为你能克服它？

Um I'm curious about first of all, why are DELs so hard? I think we just covered a lot of these. Like what in the historical community of people doing DELs was making it so traumatic as an experience. And then why did you think you could overcome that?

但我想说的一件事是，呃，当 DELs 被发明时，呃，那绝对是在测序更昂贵的时代，做测序实验感觉像是一个大的后勤和财务波动。呃，然后当你当时构建技术，然后指向它所指向的问题时，非常像是哦，我们想要找到与非常具有挑战性的靶标结合的新化学材料。所以这带来了几个战术问题。呃，其中之一是，传统上人们不会筛选大量的蛋白质靶标。他们通常感兴趣的是为非常难的东西找到化学物质。呃，他们可能尝试了一堆其他技术，但没有成功。所以好吧，让我们尝试 DEL 筛选，因为我们什么都没有。然后他们会去制造他们要筛选的蛋白质，他们会投入大量工作，因为做 DEL 筛选感觉像是一种提升。所以也许他们可能会花两到六个月生长他们关心的大批量蛋白质，进行一堆纯化。也许他们会运行一些功能测定来说服自己，哦，它真的必须这么好，我们才能去运行这个筛选。呃，然后当他们运行筛选时，他们会试图最大化获得相当好的化学材料的机会。所以激励驱动到一个地方，我们想要一个大的 DEL，就像我们想要很多多样性。所以如果你去看提供 CRO 服务的人或内部 DEL 团队，很多时候他们会推动这些非常大的 DELs，呃，希望如此，这就是他们的动机，这很有道理。我们所做的非常非常不同的事情是，呃，你知道，我们筛选成吨成吨的蛋白质，我们不太受驱动于试图为某个可能对项目高度感兴趣的特定蛋白质靶标找到单个结合物。我们只是想获得成吨成吨的活性化学材料，跨越大量蛋白质。所以在那里我们可以很好地识别我们的 DELs 的行为，因为我们的每一个化学库都被筛选针对数百个靶标。所以你更好地表征化学库。

But one thing I do want to say is that um when DELs were invented uh it was definitely at a time when sequencing was kind of more expensive and it really felt like taking doing sequencing experiments was kind of like a big logistical and financial swing. Um and then when you built the technology at the time and then pointed it at the problems that it was pointed at, it was very much oh we want to find new chemical material binding to targets that are very challenging. And so that brings up a couple uh kind of tactical issues. Um one of them is that classically folks don't screen large numbers of protein targets. They're usually interested in finding chemical matter for something that's like really hard. Uh and and and they have maybe tried a bunch of other techniques and they haven't worked out. So okay, let's try a DEL screen because we don't have anything. Then they'll go off and make uh the protein that they're going to try to screen and they'll put a lot of work into it because it feels like kind of a lift to do a DEL screen. And so maybe they might spend two to six months growing big batches of the protein that they care about, doing a bunch of purifications. Maybe they're going to run some functional assays to convince themselves that oh, it's really got to be this good uh for us to go off and run this screen. Um and then when they run the screen, they're going to try to maximize their chances of getting pretty good chemical material. And so then the incentive drives into a place where we want we want a big DEL like we want a lot of diversity. And so if you go looking at uh folks who are offering CRO services or internal DEL groups a lot of times they'll push into these very large DELs uh and hopes and that and that's their motivation and and it makes a lot of sense. What we have done that's very very different is um we you know we screen tons and tons and tons of proteins and we are less driven by trying to find uh individual binders for a particular protein target that might be of high interest for a program. We just want to get tons and tons of you know active chemical material across large numbers of proteins. And so there we can identify how our DELs behave very very well because every one of our chemical libraries gets screened against many hundreds of targets. So you're characterizing the the chemical library the chemical library better.

我们非常清楚，所以开始遇到很多这些模式，它们是我们设计的化学或制造工艺的怪癖。一个很好的例子是，我们有一个化学库，已经做了几次，即使是同一个化学库，它的表现也不同。直到我们筛选了数百种蛋白质后才发现这一点。然后你不得不盯着那些结果说，哦，现在我们看到这些模式出现得更多了。随着你对库的表征越来越多，它真的能帮助你评估和排查甚至你的蛋白质生产过程。我们觉得这总体上相当可行的一个原因是，我和公司里的几个人来自真正的基因组学方法开发背景，而不是化学背景。所以我们知道可能需要调整实验方案，才能从这些古怪的检测中获得良好的信号。我想说，那就是加上在 Recursion 的多年经验。那是我和我的联合创始人 Andrew Levens 以及我们团队中许多其他人最后待过的地方。我们都在高维生物学中犯过很多错误，在黑暗中摸索，找出可能存在的不同模式、失败模式，这些模式可能让模型在 Recursion 产生的高维成像数据上作弊。所以我们有这种真正的眼光去思考，哦，这里有这些噪声来源，我们必须追查它们。哦，你能预测这是在哪台仪器上运行的吗？如果你能，那就很糟糕。所以我们把所有这些思考都融入到追求这个特定检测中，我们感觉人们以前没有真正以那种细节水平审视过它。

We know very, very well, and so we start to encounter lots of these motifs that are quirks of the chemistry that we designed or maybe the manufacturing process. A great example is we have one chemical library that we've had made a couple times, and even though it's the same chemical library, it does behave differently. We did not know that until we screened hundreds of proteins against it. Then you have to stare at those results and say, oh well, now we're seeing these motifs show up a little bit more. As you characterize the library more and more, it really helps you evaluate and troubleshoot even your protein production process. One reason that we felt that this is pretty tractable overall is that I and a couple of the people at the company come from real genomics methods development backgrounds as compared to chemistry backgrounds. So we knew what it might take to tweak experimental protocols to really get good signal out of these funky assays. And I would say it was that plus many years at Recursion. That's where I was last with my co-founder, this guy named Andrew Levens, and a number of other people on our team. And we all made, honestly, a lot of mistakes in high-dimensional biology, bumping around in the dark figuring out what different modes, failure modes there might have been that would allow models to cheat on the high-dimensional imaging data that Recursion produces. And so we had this real eye for thinking about, oh well, you know, there's these sources of noise this way. We're going to have to chase those down. And oh, can you predict the instrument that this was run on? If you can, that's really bad. And so we incorporated all that thinking into going after this particular assay, and we had a feeling that folks hadn't really stared at it in that amount of detail before.

是的，这要追溯到 Ron Alpha noetic 那期节目，我发现——

Yeah, this goes back to the Ron Alpha noetic episode where I found—

黑手党。

Mafia.

是的，Recursion 黑手党。我发现你们两家公司之间有一些真正的精神相似之处：这种真正的认知严谨性，非常仔细地检查你所做的一切，以不同的方式和半重叠的方式做事。我只是好奇——我们问过这个——是的，非常注重数据，大规模自动化。我好奇的是：在 Recursion 是否有某种东西，就像这是 Recursion 的文化，那种——

Yeah, Recursion mafia. I found there's some real spiritual parallels between both of your companies: this real epistemic rigor, checking very carefully everything you're doing, doing things in different ways and sort of semi-overlapping ways. And I'm just curious—we asked this—yeah, very data-focused, like at scale, automation. And I'm curious: was there something like at Recursion, like was this just like Recursion's culture that is sort of—

是的，尤其是在早期。我想说有四个——我认为有四家 Recursion 黑手党公司，它们都有点押韵。有我们，有 Ron Alpha 的 Noetic，有 Vishwaluru、August Allen、Hana Gordon 和所有这些优秀的人、Dave Healey 的 Inveta，还有 Katy Rose Scaly 经营的 Known Medicine，几年前被 PAS 收购了。所以所有这些公司基本上都在研究不同的有趣生物学问题，测量极其嘈杂，然后他们从 Recursion 拉人进来，想，哦，我们很擅长处理嘈杂测量。让我们看看能否从这个特定角落获得干净信号。你知道，对我们来说是小分子化学问题，对 Inveta 来说是天然产物混合物——他们做很多质谱之类的东西——然后是 Ron 和 Katy Rose 的团队。所以是的，我们都很相似。

Yes, especially in the early years. I would say that there are four—I think there's four Recursion mafia companies and they all kind of rhyme. So there's us, there's Noetic with Ron Alpha, there's Inveta with Vishwaluru and August Allen and Hana Gordon and all these great people, Dave Healey, and then there's also Known Medicine that was run by Katy Rose Scaly that got picked up by PAS a couple years ago. And so all these companies are basically looking at different interesting biology problems with extremely noisy measurements, and then they pulled in people from Recursion to figure out like, oh, we're pretty good at noisy measurements. Let's try to see if we can get clean signal out of this particular corner. You know, and so for us it's a small molecule chemistry problem, for Inveta it's mixtures of natural products—they do a lot of mass spec kind of stuff—and then there's Ron and there's Katy Rose's group. So yeah, we're all very similar.

那么让我——我们暂停一下。你们到底做什么？

So let me—let's pause for just a second. What do you guys actually do?

太好了。抱歉，我——我太兴奋了，就插话了。抱歉。

That's great. Sorry, I like—I'm so excited about this, I just jumped in. Sorry.

太好了。

That's great.

是的，我们是一家机器学习小分子公司。这意味着我们去收集大量小分子与蛋白质靶点结合的数据。我们用它来推动机器学习对这种东西的理解的进步。我们也把小分子材料推给合作伙伴，并将其推向患者。

Yeah, we are a machine learning small molecule company. And so what that means is we go off and we collect massive numbers of small molecules binding to protein targets. And we use that to drive advancements in machine learning understanding of this kind of stuff. And we also push small molecule material into partners and move it forward towards patients.

为了不熟悉的观众，什么是小分子？我们为什么关心？它与其他疗法有何不同？

Just for the audience that's not familiar, what is a small molecule? Why do we care? How is it different than other therapeutics?

这是很棒的技术。

This is great techniques.

这很棒。

This is great.

所以，当你试图药物化某物时，我之前谈过你体内有一个感兴趣的蛋白质，你要尝试相互作用，以某种方式影响它的行为。目前市场上有几种不同的东西人们使用。常见的一种是抗体，由细胞制造。它们是大分子。所谓大，它们大到很难进入细胞内部。它们也很难做成药丸。所以如果你有抗体治疗或像 GLP-1 治疗这样的蛋白质治疗，你将不得不注射。它可以影响某些容易到达的蛋白质。但细胞内还有其他驱动许多疾病的蛋白质，对这些来说相当具有挑战性。在很多方面，与其用这个大东西——蛋白质，不如用一个小化学物质，这更有吸引力。小化学物质，这里的小分子，是可以放在药丸里的东西。它们可以进入你的细胞并影响行为。它们确实构成了大多数药物。如果你家里有任何药丸，它们几乎肯定是小分子。在那些一般性地追求治疗的不同的机器学习药物发现小组中，我们已经取得了——我们指的是你们所有人，21 世纪的生命科学家——在能够预测和实现蛋白质类型药物行为方面取得了相当好的进展。所以你有一堆抗体公司，比如 Isomorphic 做这个，或者像 Chai 或 Nava。有所有这些非常优秀、非常聪明的人在这些特定问题上推进，我认为部分原因是我们理解或有足够的数据来了解蛋白质行为，使得这些公司能够相当有效地进行这些设计。我们的论点是，这些能力对于小分子方法来说真的还不存在。所以这就是我们建立这家公司的原因之一：我们觉得我们可以以有意义的方式去追求它。

So, when you're trying to drug something, I talked a little bit earlier about how you've got a protein of interest in your body and you're going to try to interact, you affect its behavior in some way. There are a couple different things that are currently on the market that people use. One thing that's common is antibodies, which are made by cells. They're large molecules. And by large, they're big enough that they have a really hard time getting inside cells. They also have a really hard time being in a pill. So if you have like an antibody treatment or one of these protein treatments like GLP-1 treatments, you're going to have to have kind of an injection. It can affect certain proteins that are easy for them to get to. But there's other ones inside the cell that drive a lot of disease that are quite challenging for those. It is kind of more appealing in a lot of ways to instead of have this big thing that's a protein to have a little tiny chemical. And tiny chemicals, small molecules here, are things that can be in pills. They can go inside your cells and affect behaviors. They do make up most of drugs. If you've got any pills in your house, they're almost certainly small molecules. And of the different machine learning drug discovery groups that are going after treatments in a general sense, we have made—we as in all of you, life scientists in the 21st century—have made pretty good progress on being able to predict and enable the behaviors of protein type drugs. So you've got a bunch of antibody companies like Isomorphic does this or like Chai or Nava. There's all these groups of very good, very smart people pushing on these particular problems, and I think part of that is because we understand or have enough data for protein behaviors that make it reasonably effective for these companies to do these designs. Our thesis is that those capabilities are really not here for small molecule approaches. And so that was one reason why we built this company: we felt that we could go after it in a meaningful way.

而小分子，一个关键特征——它们在你体内到处走，对吧。所以这有好有坏。

And small molecules, one of the key characteristics—they go everywhere, right, in your body. So that's good and bad.

是的，它们可以穿透细胞，也可以更容易地穿过肿瘤的防御和血脑屏障。但与此同时，它们做所有这些事情，因此它们可能更容易产生毒性和其他副作用。

Yeah, they can penetrate into the cell and they can also get through defenses of tumors and cross the blood-brain barrier more easily. But at the same time they do all those things and therefore they can be more prone to toxicity and other side effects.

非常正确。

Very true.

所以小分子通过监管壁垒的路径可能比抗体更艰难一些。对于抗体，我们现在相当清楚它们在体内如何表现，它们可能引起什么问题，以及不会引起什么问题。而对于小分子，有一系列描述得很清楚的问题，其中一些可能会受到影响。例如，它们可能导致肝脏问题，也可能导致心律失常。作为一名药物化学家，其工作就是开发这些分子，把它们从在试管里看起来还行，变成你足够有信心给人类使用的东西。有很多这样的陷阱，他们必须做大量工作来避免这些陷阱。所以确保我的分子不会引起心律失常，然后如果我改变它来阻止那些问题，它是否会影响它做我想让它做的事情。你们的模型如何帮助解决这个更大的开发问题，然后也许还有这些下游问题。

So the path through the regulatory barriers for small molecules can be a little bit more fraught than those for antibodies. And so antibodies we know pretty well right now about how they behave in the body and what they are likely to cause problems and what they're not. And for small molecules there's a bunch of well-described problems that some of them might affect. They might cause liver problems for example, they might cause heart arrhythmias. And as a medicinal chemist, a person whose job it is to develop these molecules to get them from something that looks okay in a tube to something that you would feel confident enough to give to a human being. There are a lot of these pitfalls and they have to do a lot of work to avoid these pitfalls. So make sure that my molecule doesn't cause heart arrhythmias and then if I change it to stop that's causing those problems does it affect it doing the thing that I want it to do. How do your models help with this larger problem of developing these and then maybe with these downstream problems.

所以一种思考方式是，其中一些问题，这些所谓的脱靶毒性，将是由你的小分子以你可能能够预测或识别的方式结合其他蛋白质所驱动的。因此，这是我们正在大力推进的事情，就是让我们也制造那些蛋白质。让我们也用我们的化学方法筛选那些蛋白质。

So one way to think about it is that a handful of these problems, these sort of off-target toxicities, are going to be driven by your small molecule binding to other proteins in ways that you might be able to predict or identify. And so that's something that we are very much pushing forward to is oh let's also make those proteins. Let's also screen those proteins with our chemistry.

如果你有一个巨大的检测，那似乎很有用

Seems useful if you have a giant assay of

没错。

That's right.

不同蛋白质的检测，以及一个你可以

of different proteins and a library that you can

是的。毫不夸张地说，我们的愿望是筛选每一种蛋白质，然后也扩展到人类之外的蛋白质

Yeah. It is not at all an exaggeration to say that it is our aspiration to have screened every protein and then also move on to those outside of humans

既是为了获得蛋白质的变异，也是为了识别也许你不想运行小鼠模型，也许你应该运行大鼠模型

both for reasons to get variation in the proteins but also to identify maybe you don't want to run a mouse model maybe you should run a rat one

因为你的化学物质在不同物种中行为不同，因为这些蛋白质可能略有不同。

because your chemical material behaves differently in those different species because those proteins might be slightly different.

有趣，

Interesting,

对吧？所以你想做这些事情。但是的，追求大量这些事物让你能够思考，好吧，现在我有我的化学物质，然后我知道所有这些化学物质中，哪些以及哪些变体或可能结合这些物质的会导致其他毒性问题。

right? So you want to do that stuff. But yes, going after large numbers of these things allows you to think, all right, well, now I've got my chemical material and then I know of all these chemicals, which ones and what variations or ones likely to bind these things will cause other toxicity problems.

明白了。

Got it.

是的。是的。然后在那之后，有许多其他人做的其他检测，这些检测更像是涌现行为，可能不是单个蛋白质，而是它如何很好地穿过细胞膜，

Yep. Yep. And then downstream of that there's a bunch of other assays that lots of people do that are more kind of emergent behaviors where it's maybe not a single protein, but it's how well it gets across cell membranes,

它如何以更复杂的方式相互作用，需要细胞来评估。或者可能像心脏细胞，例如用于这种心律失常的东西。所以这些事情，你知道现在有很多团体正在推动在细胞水平上测量这些的技术，这是你知道的，离我们现在关注的东西差一个生物抽象层次。但我们喜欢认为，对生物物理学本身有良好理解的模型可能有用，也许将来可以在较小的数据集上进行微调。

how it interacts in a more complex way that you require a cell to evaluate. or it might be like a cardiac cell for example for this heart arrhythmia stuff. And so those things you know right now there are lots of groups that are pushing on technologies to measure these at the cellular level which is you know one level of biological abstraction away from what we're focusing on right now. But we like to think that models that have a good understanding of the biophysics themselves can be useful and maybe fine-tuned on smaller data sets in the future.

也许再往回跳一点，稍微进入模型。为什么是 DEL？DEL 至少在我们看来，确实是测量小分子相互作用的最可扩展技术，因为你可以把数百万个放在一个管子里。现在，关于这些化学库，我们还没有谈到的一点是，当你以混合过程制造它们时，可能会有一些反应不完全，你认为你有的标签翻译可能有点偏差。所以我们在构建化学库时设计了很多对照来缓解这个问题。当你使用这些，你知道，可能叫阵列库，或者一次一个化学物质在管子里或板孔里的库，当人们研究以弄清楚这些标签有多好时，你希望它们很好，但当人们检查时，它们通常偏差 20% 到 30%。这是一个巨大的错误标记问题。我们认为这可能与 DEL 中的错误标记相当，也许更糟。而 DEL 本身只是非常非常可扩展。也许我可以，我会讲一个快速的故事，我相信我们会在某个时候谈到。总之，我们几年前举办了一个 Kaggle 竞赛，我们决定为 Kaggle 竞赛生成数据并发布它。有一个现有的，有一个数据库，两年前叫 BindingDB，它是小分子与蛋白质靶点结合的最大聚合集合，与细胞或其他东西相比。它大约有两，略低于三百万个物理相互作用。所以我们发布了一个大约四十亿个物理相互作用的数据集，而且它全部是由一个人在我家地下室测量的。

Popping back one more maybe and getting a little into model. Why DELs? DELs really are at least in our view the most scalable technology for measuring small molecule interactions because you can put many millions of them in a tube. Now one thing that we haven't touched on in terms of like these chemical libraries is that when you make them in a pooled process there are probably going to be some that react imperfectly and that the label translation of what you think you have might be a little bit off. And so we've designed a lot of controls into constructing our chemical libraries to mitigate this. When you're using these, you know the word might be array libraries or libraries that are in one chemical at a time in a tube or in a well in a plate, when folks study to figure out how good those labels are like you hope that they're good but when folks examine this they're typically off by 20 to 30%. It's an enormous mislabeling problem. We think it's probably comparable to mislabeling in DELs maybe worse. And DELs themselves are just enormously enormously scalable. And maybe I can, I'll tell a quick story that I'm sure that we'll get into at some point. Anyway, we ran a Kaggle competition a couple years ago and we decided to generate the data for the Kaggle competition and we released it. There's existing, there's a database that was at the time two years ago called BindingDB which was the biggest aggregated collection of small molecules binding to protein targets as compared to cells or something else. And it was about two, a little under three million physical interactions. And so we released a data set that was about four billion physical interactions and it was all measured by a single person in the basement of my house.

是的。早期的博客文章真的很有趣。你有一张字面意义上的照片，你有一个行动。这不太像车库创业，但在地下室。

Yeah. The early blog posts are really funny. You have a literal picture you have an act. It's not quite a garage startup, but it's in a basement.

这是一个地下室公寓，天花板大约 7 英尺高。是的。

It is a basement apartment with a like a 7 foot ceiling. Yeah.

看起来幽闭恐怖。是的。我从未见过生物技术创业公司在地下室开始。

It looks claustrophobic. Yeah. I've never seen a biotech startup started in a basement.

它会像那样完成，对吧？

It'll be done for like that, right?

但好吧，所以这个 Kaggle 竞赛我认为只是一个迷人的故事，因为你进入这个时抱有如此高的希望。

But okay, so this Kaggle competition I think is just a fascinating story because you had such high hopes going into this.

我们确实有。

We did.

你就像，我认为这是那种非常不错的认识论严谨性的案例之一，你没有只是提供哦，就像你知道这里有一个随机分割的训练测试之类的。你仔细思考了如何组织这个 Kaggle 竞赛，如何运行它以及如何公平地评估。你能谈谈那个过程，你做出的决定，然后还有结果吗？

You were like and I think this is one of those cases of like very nice epistemic rigor where you didn't just provide oh like you know here's a train test with random split or something. you thought carefully about how to organize this Kaggle competition, how to run it and how to kind of evaluate fairly. Can you talk about that process, the decisions you made and then also the result?

是的，很乐意。所以我们远不是第一个尝试评估你是否可以使用计算机对小分子与化学物质结合做出合理预测，然后测试这些预测的团队。有很多论文这样做，比如有一个真正的经典论文，出自 XCM 和 Google，在我们创立 Leash 几年前就做了这个。我们在公共数据上运行了很多自己的实验。我们怀疑没有人真正擅长这个。所以我们想把这个竞赛放在一起，并特别设计评估，以真正测试它在一系列评估难度上的表现，因为我们在想，好吧，我们将有很多人，希望很多人参加这个竞赛。有些人会非常优秀，有些人可能是初学者。

Yes, happy to do that. So we were not by a long shot the first group to try to evaluate whether you could use computers to make reasonable predictions of small molecules binding to chemicals and then test those predictions. There are lots of papers that doing this like there's a real classic paper that came out of XCM and Google some years before we founded Leash that does this. We had run a lot of our own experiments on public data. And we had suspected that no one was really very good at it. And so we wanted to put together this contest in a and design the evals in particular in a way that could really test it across a range of eval difficulties where because we were thinking about okay we're going to have a lot of people hopefully a lot of people compete in this. Some are going to be very good, some are probably going to be beginners.

We want to make sure that there are some nice easy wins, then a bunch of stuff in the middle that's kind of hard, and then some things at the very top end of what would be difficulty or what we imagine difficulty might be. To unpack that, we've talked a little bit already about how these libraries are combinatorial. So you're going to have some smaller number of building blocks, but a large number of combinations. We got our hands on a chemical library from a wonderful sponsor, this guy named Shiaoji Louu who runs a shop called Alpha Mal in Shanghai. The chemical library was about 130 million molecules. The center part of the molecule, you can imagine it kind of like a hub with three spokes, and the center hub was the same for every molecule. It's a particular chemical motif called a triazine. So what we did was we said, "Okay, we got the center hub. Then we got these three spokes for the evaluation data set. Let's have a couple different ways to split this." One way is that we could say, "All right, the chemicals might have one shared building block, one shared spoke, and then the other two spokes are different." Okay, that's one thing we could do. So let's do that for some. And that was the easy stuff. Another one we can say is, all right, well maybe we're not going to have any shared building blocks, but we do know that if they're going to come from this chemical library there's going to be the shared hub, and so we can imagine that some of the spokes might be similar but they won't be exactly the same and they won't be shared. So we think this is going to be a little bit harder. Sounds great. And then for the third most difficult eval, we had a completely different chemical library that we had designed at Leash. There was no common hub, no common building blocks, different attachment chemistries, different everything. So we thought, okay, we're going to have these three in different proportions. We're not going to tell anyone that's what we did. We're just going to say here are the evals and let's go see how it goes. So we launched the competition. It was two years ago. You guys can look it up if you want. We've written a lot about it. And then after 2,000 teams competed, the winner did well on the stuff where there were shared building blocks. Depending on the target, we saw some boosts in performance for where there were no shared building blocks, but the core hub was still the same. The triazine was still there. And then when you go to the most challenging eval, no one did better than chance.

Yeah. The winner was like the 1,200th just random person who submitted a Jupyter notebook, right? Or something.

Actually, yeah. So the winner was this guy named Victor Schlapov who put together a slightly different approach, but his was good, but again, like everyone else, no better than chance. Three of the top five were the noisy tutorial notebook. So, but just to be clear, there usually with a Kaggle competition, there's a public eval for you to train to test your algorithm, I guess, and then there's a private one that's held out. That's right. And the competition runs that and all you know is the score.

So, was your library included in the public?

It was hidden back to make it spicy.

Right? Okay. Yeah.

And it was spicy.

It was very spicy. So what did you take from this? What was the lesson? And first of all, what did you hope to get from this? You kind of touched about that, but then how did that affect your thinking about how do you proceed forward there?

Yeah. We had a couple hopes and some fears. One thing I will be candid about is we had released this data set that was a thousand times bigger than anything anyone had ever seen. And we gave it about a 5% chance that recent advancements and kind of like coalfolding approaches would mean that it would obviate our company entirely. Like didn't happen.

No it didn't happen.

No, but we were like, you know, there's like a small chance like we are really messing up here and that didn't happen. We suspected more likely that folks weren't very good at it and that the data might help and so we're hoping to see that. One thing that—

And anyway, you'd want to know that.

You'd want to know that before you 10 years of your life.

Absolutely. Absolutely.

And another thing that we wanted to know is there's lots of different ways to represent molecules and give them to computers. You could use these things called SMILES that do it some way. You can do 3D graphs. There's like weird spherical convolutional nets. People have tried kind of goofy things to try to represent molecules in a useful way. And there'll be lots of people in the field who will argue that, oh I think this way is better than this way or whatever. And we did really encourage competitors, especially those who had a chemistry background, like let's try some of these different representations and let's see how it goes. And we were grateful to find that the thing that we were doing, which are just basic SMILES, kind of beat everything else. It wasn't like a resounding thumping, but it was better enough for us to continue using that approach.

And SMILES just, if I recall correctly, is basically a sequence, a textual sequence of strings. It doesn't encode any 3D structure into it. It's all emergent from—

It is a graph traversal, yeah.

Translated into a text string. And so you can start the graph at different points. And so there's what they call canonical and non-canonical SMILES depending on where you start. But yeah, it's an extremely simple representation.

And so yeah, we were thinking that, you know, we wondered if some of these fancier ways of representing molecules might beat at least this particular problem and they super didn't.

So was your takeaway that you should go full force into this? Was this encouraging to you or was this discouraging you? What was the collective emotional reaction at Leash after that? I always wondered that. I mean he kind of wrote about you had several follow-ups. But I'm just curious to hear.

Yeah. Yeah. This is a great question. So like we are extremely bitter lesson pill over here. And so our thought was—

Some of it was kind of delight which is like wow this is much harder than we thought. That's kind of great.

That's kind of great because if this problem is like all the other ones that we're aware of.

如果你能从一个可扩展的测量体系中获得相当不错的信号，那你就应该直接去争取。

If you can get pretty good signal out of a scalable measurement regime, then you should just try to go get that.

是的。是的。不，我的意思是，对我们来说这相当有验证性，就像，天哪，这里有个机会。

Yeah. Yeah. No, I mean, for us it was pretty validating, like, oh man, this is an opportunity here.

尤其是如果你的可扩展测量体系很难，而你有专门的知识来做这件事。

Especially if your scalable measurement regime is hard and you have specialized knowledge to do that.

绝对。绝对。就像，天哪，是时候在这上面浇油了。

Absolutely. Absolutely. It was like, oh man, this is the time to pour gas on this.

这太棒了。

This is great.

哇。是的。我的意思是，那太棒了。所以然后你就说，好吧，全力推进。我们就开始收集数据，收集数据，收集数据。

Wow. Yeah. I mean, that's awesome. So then you were just like, okay, let's full force. We're going to just start collecting data, collecting data, collecting data.

没错。

Correct.

是的。所以你这样做了一段时间，然后接下来发生了什么？也许首先，你在过程中学到了什么教训，然后构建出你第一个有用的成果？

Yeah. And so you did that for a while, and then what came next? And maybe first also, what were the lessons you learned in the process, and then building to your first useful outcome?

是的。是的。从过程中我们觉得有几件事非常非常有用。我们意识到的一件事是，那个 count 竞赛是基于三个研究得相当透彻的蛋白质靶点。我们很清楚，出于很多原因，包括这些，要在新蛋白质上获得性能可能意味着我们得做更多得多的工作。所以，好吧，我们一直有动力这样做，但这真的凸显了，天哪，我们得去，我们得制造很多蛋白质，对吧？所以这是其中一部分。另一部分是认为我们有这个非常大量的小分子，比如这个 1 亿化合物库，但对我们来说，它不如更好地表征一个更小的库有帮助。所以这也是我们做出的一个战略选择。这其实相当重要，因为你有一个固定的预算，所以你可以要么，比如说，在测序上花更多钱，为你已经有的更小的集合获得更高的分辨率，要么你可以获得更多分子，然后希望噪声会平均掉之类的。

Yeah. Yeah. There are a couple things that were from the process that we thought were very, very useful. One thing that we realized is that the count competition was based on three protein targets that were fairly well studied. It became clear to us that we really wanted to, for lots of reasons, but including these, that getting performance across novel proteins probably meant that we'd have to do a whole lot more. So, okay, so we were always motivated to do that, but that really underscored like, oh man, we're going to have to go, we're going to have to make a lot of proteins, aren't we, here? And so that was part of it. Part of it was thinking that we had this very large number of small molecules, like this 100 million compound library, that didn't help as much to us as better characterizing a smaller one. And so that was also a strategic choice that we made. This is actually quite important because you have a fixed budget, and so you can either, let's say, spend more money on sequencing, getting higher resolution for what you already have for a smaller set, or you can get more molecules and then hope that the noise sort of averages out or something.

但这是正确的。就像拥有固定预算确实会影响你的战略选择。对我们来说，我们想确保我们在两个轴上都有很多变化，而不是一个轴和少数几个蛋白质。

But that's correct. Like having a fixed budget really does affect your strategic choices. For us, we wanted to make sure that we're getting a lot of variation across both axes as compared to like one axis and a handful of proteins.

抱歉，哪两个轴？

Sorry, which two axes?

那就是化学品的数量对蛋白质的数量。好的。对。所以，你可以想象，比如，哦，我要在 10 亿化学品中做三个蛋白质，或者也许我应该做数千个和个位数百万。是的。或者这些的其他组合。

So that would be like the number of chemicals against the number of proteins. Okay. Right. And so, you know, you can imagine like, oh, I'm going to do three proteins in a billion chemicals or maybe I should do like thousands and single digit millions. Yeah. You know, or some other combination of those.

个位数百万可能是曾经构建的最小的细胞之一。

Single digit millions being probably one of the tiniest cells ever constructed.

它相当小。我见过的最小的是 5 万分子。

It's pretty small. The smallest one I've seen is 50,000 molecules.

哦，好的。是的。我想如果你想做类似 S 的东西，你可以定制一个 Dell。

Oh, okay. Yeah. I guess you can make a custom Dell if you want to do like S.

你可以做任何你想做的事。

You can do whatever you want.

是的。是的。是的。当你有技术时，就去做。两个构建块。是的。

Yeah. Yeah. Yeah. It's a when you have the technology, just do it. Two building blocks. Yeah.

是的。对。你可以做任何你想做的事。所以好处是你对库的大小、特性以及我提到的条形码问题都有巨大的控制权。你可以扩展这些东西。我们在自己的 Dell 中投入了很多思考，比如各种花哨的功能，以真正提升性能和信噪比等等。但回到 CG 竞赛的教训，我们也清楚地认识到，模型可能会考虑记忆单个构建块，这会提升它们的性能。这似乎很清楚。它们可能通过记忆获得一个想法，即如果你有一个非常相似的构建块，它可能表现相似。对于某些性能指标，以及我们有一些相当远的东西，我的意思是，在当前体系下，在那个体系下它们根本做不到。当我们发布这些结果时，基本上说嘿，没有人比随机做得更好，我们确实得到了一些反弹，人们说，好吧，有没有优秀的人参加你的竞赛？也许每个人都很差，也许都是一年级研究生。所以结果我们发布了所有数据，全部。所以如果外面有人认为自己很棒，请展示给我们看，因为我们想解决这个问题。但我们还没有看到有人在那上面做得很好。

Yeah. Right. You can do whatever you want. And so that's the nice thing is you have enormous control over both like the library size and its characteristics and like this barcode issue that I talked about. You can expand that kind of stuff. And we put a lot of thought into kind of bells and whistles in our own Dells to really boost kind of like a performance and signal over noise and all this stuff. But getting back to the kind of like the CG competition lessons, you know, it was also clear to us that models could think about memorizing individual building blocks and that would boost their performance. That seemed pretty clear. It seemed pretty clear that they might be able to get an idea through probably through memorization that if you have a very similar building block, it might behave similarly. To also some of those performance metrics and then also that we have something that's pretty far away, they, I mean, at the current regime at that regime they simply could not do it and when we released these results basically saying hey no one did better than chance we did get some blowback where folks were like well did anyone good compete you know in your competition like maybe everyone's just bad maybe it's all first year grad students and so as a result we just released all of the data entirety and so if anyone out there thinks they're awesome then like please show us cuz like we want this problem solved. But we haven't seen anyone do great perform great on that stuff yet.

当你说所有数据，包括所有原始读取计数和一切。所以你可以仔细检查并说哦，这不仅仅是处理错误，你可以从原始数据开始使用它。

And when you say all the data you have including all like the raw read counts and everything. So you can go through and say oh it's not just a processing error like you can go from the raw data and use it.

你可以拿，是的，你可以拿我们的二元标签，你可以自己做。所有的评估，一切，都在 Polaris 上，那是一个由 Recursion 运营的非营利组织。它在那里供任何人使用。请使用它，我们故意发布它，希望社区能好好利用它。

You could take, yeah, you could take our binary labels, you could do it yourself. All the evals, everything, it is all on Polaris, which is like a nonprofit group that was run by Recursion. It's there for anyone to use. Please use it, like we intentionally released it as, you know, hoping the community would make good use of it.

不过，你认为这个数据集的结构在某种程度上根本上是有限的，也许没有一些外部增强，你无法做得比人们已经做的更好。

Do you think though the structure of this data set was fundamentally limited in a way that maybe without some external augmentation, you're not going to be able to do any better than what people have pretty much already done.

我认为如果你有真正完美的结构理解，或者非常好的结构模型，那么你应该能在这上面做得不错，对吧？因为如果我们都有模型真正理解小分子氨基酸之间相互作用的物理，那么如果它们真的很擅长这个，小分子训练集的分布是什么样就不重要了。我认为当我们看到结构模型预测时，人们说，哦，好吧，你知道，我在 PDB 上有一个时间分割，例如，你知道，我们的模型从未见过这些分子或这个蛋白质，然后希望我们能利用这种理解来做出合理的折叠。这非常符合能够对这些东西做出预测的精神，因为没有人见过这些，尤其是我们的内部小分子，没有人见过那些。

I think if you had really perfect structural understanding or like really good structural models, then you ought to be able to do okay on this, right? Because if we all have models that really understand truly understand like the physics of the interactions between the amino acids of the small molecules, then it doesn't really matter what the distribution of the small molecule training set would look like if they're really good at this, you know, and I think when we see when we see structural model predictions and folks say, "Oh, well, you know, I've got this time split on the PDB, for example, and you know, you know, our model's never seen these molecules or this protein before, and then hopefully we can, you know, kind of use this understanding to make reasonable folds." Like that's very much in the spirit of being able to make the predictions on this stuff cuz no one has seen these especially our internal molecules like no one's seen those before.

那很酷。所以 Kaggle 竞赛你继续前进，你学到了一些教训。你全力收集数据，重新集结，致力于 Scaling。

That's cool. So Kaggle competition you went ahead and you've learned some lessons. You went full force data collecting regrouped committed to scaling.

是的。

Yep.

好的。那么然后你，我想大约六个月后，你发布或有一篇关于你的模型 Hermes 的博客文章。

All right. So then you I think something like six months later you released or you had a blog post about your model Hermes.

是的。

Yep.

你能给我们讲讲那个吗？还有其中的一些经验教训，特别是你如何评估那个模型，以及你投入的所有心血？

Can you tell us about that, and some of the learnings there, and especially how you evaluated that model and all the care you put into it?

Hermes 是一个模型，最初是由 Le 的一位数据科学家 Mike Kleiner、Max Kleinesser，还有 Sean Francis Line 构思的，当然 Andrew Bleven 是那里的 CTO。最初的想法是，我们有一堆数据，我们在考虑如何划分数据以及不同的训练方案，所以我们想出一个非常非常简单的模型，这样我们可以进行大量潜在的实验迭代。它的目的不是成为最先进的模型，而是像，我们想快速尝试东西。所以我们构建了 Hermes，然后 Hermes 的基本架构是两个不同的嵌入。一个是进化尺度建模，ESM 嵌入——那是一种蛋白质嵌入，来自一个曾在 Meta 的团队，然后他们离开去创办了自己的公司，最近被 Chan Zuckerberg 收购了。

Hermes is a model that is the original brainchild of a data scientist at Le named Mike Kleiner, Max Kleinesser, and also Sean Francis Line, and Andrew Bleven, of course, is the CTO there. The thinking originally was that we had a bunch of data and we were thinking about how to split it and different training regimes, so we wanted to come up with a model that was really, really dead simple so that we could do lots of potential experimental iterations. Its purpose wasn't to be a state-of-the-art model; it was like, we want to try stuff fast. So we built Hermes, and then the basic architecture of Hermes is two different embeddings. One of them is an evolutionary scale modeling, ESM, embedding—that's a protein embedding that comes from a group that was at Meta and then went off to start their own company, and then they got picked up by Chan Zuckerberg more recently.

我们播客请过 Alex Reeves，所以如果你想了解更多，可以回顾那一期——回到几年前。是的，谢谢你，Alex，如果你在听的话。

We had Alex Reeves on the podcast, so we can throw back to that episode if you want to hear more—go back a few years. Yeah, thank you, Alex, if you're listening.

所以我们在蛋白质方面用了那个。然后我们有一个不同的嵌入叫 Kvera，基本上就像是一个 X 令牌预测器化学之类的东西在另一边。

So we did that for the protein side. And then we had a different embedding called Kvera, which is basically like an X token predictor chemistry kind of thing on the other side.

预测 SMILES 的东西还是——

Something which predicts SMILES or—

没错。是的。就是预测 SMILES 令牌。就这样。这两个非常非常简单的嵌入，然后我们用交叉注意力把它们放在一起，就这样。然后我们用它来预测一个二元标签。

That's right. Yeah. Just predicts SMILES tokens. And so that's it. These very, very two simple embeddings, and then we just put them together with cross attention, and that was it. And then we used it to predict a binary label.

最简单的东西。

Simplest possible thing.

它非常简单。非常非常小，然后我们在上面训练了一些数据。

It's incredibly simple. Very, very small, and then we trained some data on it.

其实，你知道，不是最简单的。你可以在那些之后做一个 MLP，对吧，而不是交叉注意力。那为什么用注意力？

Actually, you know, not the simplest. You could have done an MLP after those, right, instead of cross attention. So why attention?

注意力的思考与我们之前在蛋白质结构的图卷积类图模型上的一些其他经验有关。我们注意到的一件事是，如果你有这类图，并且在我们做过的其他工作中，有几个模型这样做，我们打乱了那些的位置坐标，而其中一些模型的性能或多或少保持不变。这让我们认为注意力非常关键。所以我们最终只是确保模型至少——至少我们希望的是，它会查看蛋白质的不同部分到小分子结构，这可能在这里有用。

The thinking with the attention was had to do with some other experiences that we'd had with graph convolutional-like graph models for protein structure. And one thing that we noticed is if you've got these sorts of graphs and there's a couple models that do this at other work that we had done, we had scrambled the positional coordinates of those and performance for some of those models more or less stayed the same. And that made us think that attention was super critical. And so we ended up just making sure that the model was at least—at least what we hoped is that it would be looking at the different parts of the proteins to small molecule structures that could be useful here.

所以这就是我认为在图像生成中经常被称为毕加索问题的情况，比如说，CNN 曾经出名一段时间，你可以打乱图片的特征——像毕加索那样，比如把鼻子翻转过来或者旋转眼睛之类的——而 CNN 仍然会描述，哦，这是毕加索，因为它只是捕捉到了个别的局部特征，而不是某种全局结构。这在小分子药物发现中已经是一个问题一段时间了，它经常捕捉到局部基序，然后对拓扑结构没有任何全局理解。所以特别是对于像 DEL 这样的东西，你有单个构建块的组合结构，你可以想象模型独立学习每个构建块，然后基本上只是分解为三个构建块的总和之类的。我的意思是，这是一个简单的概念。

So this is, I think, oftentimes known as the Picasso problem in image generation, where, let's say, CNNs, for example, were famous for a while where you could scramble just the features of a picture—like Picasso, like flip the nose around or just rotate the eyes or something—and the CNN would still depict, oh, this is a Picasso, because it just picked up on individual local features rather than some sort of global structure. And this has been a problem in small molecule drug discovery for a while, where oftentimes it picks up on local motifs and then it doesn't have any global understanding of the topology. So especially for something like DELs, where you have this combinatorial structure of individual building blocks, you could imagine the model learning each building block independently and then essentially just decomposing into the sum of three building blocks or something. I mean, it's a simple concept.

是的。是的。所以交叉注意力在某种意义上是强制某些分子与蛋白质中的某些残基相互作用。

Yeah. Yeah. So cross attention in some sense forces certain molecules to interact with certain residues in a protein.

是的。或者至少可能不是强制，而是提供机会。想想看。这是一种归纳偏置。

Yep. Or at least maybe not forces but provides the offer. Think about it. It's an inductive bias.

是的。你大概想说的是，你知道，我们在考虑如何评估这个，所以我们做了一些类似的技巧，就像我们在 Kaggle 竞赛中做的那样，我们有内部生成的训练集和数据，然后我们查看了更远的化学库。但然后我们还想考虑划分来探测 Hermes 的行为。因为 DEL 有这种组合方法，你可以做同样的事情，你可能会共享某些构建块。所以我们这样做，当然性能更好，我们怀疑这是由于记忆。但另一件事是，你可以想象有些构建块彼此非常相似，如果你只是试图根据它们的身份来划分，那么你可能会得到足够接近的东西，记忆或其他作弊模式可能会转换预测能力。所以我们想出了一个方案，然后我们基本上对这些不同的构建块进行聚类。我们有基于相似性的构建块聚类。这样我们就可以避免整个聚类在测试和训练之间重叠。所以当你这样做时，你知道性能确实会下降。

Yeah. Kind of what you're getting at is, you know, we were thinking about how to evaluate this, and so we had done some of the similar tricks that we did over with the Kaggle competition, where we had training sets and data that we generated internally, and then we'd looked at chemical libraries that were further away. But then also we wanted to think about splits to kind of probe Hermes' behavior. And so because DELs have this combinatorial approach, you can do the same thing where you might share certain building blocks. And so we do that, and of course performance is better, which we suspect is due to memorization. But the other thing is you can imagine that there are building blocks that are very similar to one another, that if you're just trying to split on the identity of those, then you might have things that are close enough that memorization or some other cheat mode might convert predictive power. And so we came up with a scheme where we then did basically clustering of these different building blocks. We had building block clusters on similarity. And that way we would just avoid entire clusters from being overlap between test and train. And so when you do that, then you know the performance does drop.

所以你可能会提到的一件事是，我们有——你知道，我们在这里使用了很多合理的基线，我们真的很担心。所以对于我们做的几乎所有事情，我们都会使用像 XGBoost，对吧？你应该使用类似的东西来看看进展如何。

So one thing that you might be alluding to is we have—you know, we use lots of reasonable baselines here that we're really worried about. And so for almost everything that we do, we'll use like XGBoost, right? You should use something like that just to see how it's going.

对。是的。每个——是的。经典的年轻机器学习实践者，他们花六个月优化他们最喜欢的模型，然后有人走过，做了像 XGBoost、随机森林，他们就像，“哦是的，那刚刚打败了六个月的工作。”哦是的。是的。

Right. Yeah. Every—yeah. The classic young machine learner practitioner, they spend six months optimizing their favorite model and then someone walks by and does like XGBoost, random forest, and they're like, "Oh yeah, that just beat like six months of work." Oh yeah. Yeah.

是的。所以我们做了这些不同的划分方案，试图评估事情相对于实际提升性能会做得如何。

Yeah. And so we did these different split regimes to try to evaluate how well things would do relative to like actually boost performance.

所以一件非常非常令人欣慰的事情是，你看到 XGBoost 随着你通过这些构建块聚类方法越来越远而稳步下降。就像如果你共享两个构建块，它处于某个水平。如果你只共享一个，它会变差。如果你共享一个，你知道，只有没有，它更差。如果你做构建块，它比那更差。然后我们将其与 Hermes 的性能进行比较，后者在所有这些不同的方案中相当稳定。

And so one thing that was really, really gratifying to see is that you see XGBoost drop steadily as you get further and further away with these building block cluster approaches. It's like if you share two building blocks, it's a certain level. If you only share one, it gets worse. If you share one, you know, only none, it's worse. And if you do building block, it's even worse than that. And then we compared that to the Hermes performance, which was pretty steady across all these different regimes.

我明白了。

I see.

是的。所以这让我们觉得，哦是的，你知道，你真的必须相当远离分布才能让 XGBoost 真正崩溃。这是我们没有——你知道，我们没有——它在几个图表和我们发表的一些工作中。我们没有过多地大声宣扬。

Yeah. So it made us think like, oh yeah, you know, you've really got to get pretty far out of distribution for XGBoost to really tank. And that's something that we haven't—you know, we haven't—it's in a couple plots and some work that we published. We haven't yelled about it too much.

等等，我——我以为——你不是有一篇博客文章叫 XGBoost 就是你所做的一切吗？这就是为什么我们这样做。

Wait, do I—I thought—don't you have a blog post like XGBoost is all you do? That's why we did that.

是的。

Yeah.

运行那些拆分是我们花了很多心思去思考的事情，试图弄清楚，好吧，你怎么阻止你的模型真正记住和记住东西？这可能是一个很好的时机来稍微绕个弯，但引用你另一篇博客文章，我从自己的个人经历中发现它绝对搞笑。我见过机器学习模型失败，比如捕捉到许多奇怪的干扰因素，而你关于分子所有权或作者身份的博客文章是一个非常有趣的故事。

Running those splits is something that we've put a lot of thought into, trying to figure out, okay, how do you stop your models from really memorizing and remembering things? This might be a good time for a little bit of a detour, but referencing another one of your blog posts which I found absolutely hilarious from my own personal experience. I have seen machine learning models fail, like pick up on so many weird detractors, and your blog post about molecule ownership or authorship was a really fun story.

《聪明汉斯》那篇。是的。很多年前，我和我的联合创始人安德鲁·布莱文斯在 Recursion 公司，看着屏幕上的一个小分子，然后一位药物化学家从我们身边走过。这家伙叫迈克·卡雷西。迈克说：“嘿，看起来像斯图尔特·特里伯的分子。”然后我们就像，“哦不，你只要走过就能看出这是谁做的？太可怕了。”然后这件事尤其困扰了安德鲁很多年。大概有六七年。我们开玩笑，你知道，每隔几个月我们出去喝酒时就会说，天哪，我们真的应该弄清楚这个问题有多严重。

The Clever Hans post. Yeah. So many years ago, my co-founder Andrew Blevens and I were at Recursion looking at a small molecule on a screen, and then a medicinal chemist walked by us. This guy named Mike Karesi. And Mike was like, "Hey, looks like a Steuart Triber molecule." And then we were like, "Oh no, you can tell just by walking by like who made this? That's horrifying." And then that haunted Andrew in particular for like many years. Like it was like six or seven years. And we joked, you know, I don't know, every couple of months we're out for beers like man we really should figure out like how bad this problem is.

为什么那是个问题？

Why is that a problem?

好吧，问题在于很容易想象你会尝试训练模型来理解化学。所以你会从很多小组获取数据，因为从很多小组获取数据非常诱人。如果模型仅通过观察分子就能识别化学家，那么也许模型可以推断化学家的意图，这意味着也许它们根本学不到任何东西。所以，

Okay so it's a problem where it's very easy to imagine you're going to try to train models to understand chemistry. So you're going to get the data from lots of groups because it's very tempting to get data from lots of groups. And if the models can identify the chemist by looking at the molecule alone, then maybe the models can infer chemist intent, which means maybe they don't learn anything at all. And so,

嗯，它们确实学到了一些东西。

well, they did learn something.

它们学到了一些东西。是的。是的。它们没有学到物理，对吧？它们没有学到你试图让它们学的东西。所以我们真的很害怕这个。

They learn something. Yeah. Yeah. They don't learn physics, right? They're not learning the thing you're trying to get them to learn. And so we were really afraid of this.

它们学到了某些化学家比其他化学家更好。

They learned that certain chemists are better than other woods.

是的。我们学到了各种各样的东西，但也许为观众设定一下。所以我们出去，我们在圣诞节期间发表了这篇。你可以查一下。我们有一个排行榜。我们有一个愚蠢的在线游戏你可以玩。但我们从 Kemble 下载了所有化学家的所有化学品，这是这些东西及其活性的最大存储库。每个发表超过我认为是 30 篇论文和 600 个分子的化学家。所以你得到所有分子和所有东西。有 1,815 位化学家，然后我们训练了一个分类器，仅通过观察分子来预测化学家，我们得到了 60% 的前五准确率。这是一个非常非常容易的问题。

Yeah. We learned all kinds of stuff, but maybe to set this up for the audience. So we went out and we published this over Christmas. You can look it up. We have a leaderboard. We have a goofy online game you can play. But we downloaded all chemicals from all chemists at Kemble, which is the biggest repository of these things and their activities. Every chemist who published more than I think it was 30 papers and 600 molecules. So you get all molecules and all the stuff. It's 1,815 chemists and then we train a classifier to predict a chemist from looking at the molecule alone and we got a top five accuracy of 60%. It's a very very easy problem.

是的。

Yeah.

是的。模型做起来超级容易。它们困惑的地方是当你有两位化学家在同一篇出版物上并且在同一机构。所以我们认为很可能的是，个人在制造分子时有一种独特的化学签名风格。然后当这些人有学生或实习生时，他们会采用其中的一部分，你知道，所以这对我们来说似乎非常清楚。所以这本身就足够可怕了。这是一个许多学术项目、许多博士训练过、试图预测、寻找模式的数据集，这真的是该领域的一个主要数据集。

Yeah. It's super easy for the models to do. Where they get confused is where you have two chemists who were on the same publications and were at the same institution. And so we think it's very likely is that there is a distinctive chemical signature style that individuals have when they make molecules. And then when those individuals have students or trainees, then they adopt some fraction of that, you know, and so that was that seemed pretty pretty clear to us. And so that was horrifying enough on its own. And this is a data set that many many academic projects many PhDs have trained on, tried to predict on, find patterns and this is really like a major data set for the field.

是的。就像如果你不生成自己的数据，你就在使用这个。然后如果你在生成自己的数据，那么你可能也是这个的受害者，因为你的组织可能有多个化学家，他们有自己的独特化学风格，并且他们使用自己的风格去追求不同类型的项目。

Yeah. Like if you are not generating your own data, you're using this. And then if you are generating your own data, then you are probably victim to this because your organization probably has multiple chemists who have their own distinctive chemical styles and that they are using their styles to go after different types of projects.

但这里的关键我认为是那些化学家也倾向于解决类似的问题。就像我可能在这里有一个职业生涯，我有一种风格，但我也在从事一类问题，对吧？所以你的风格也是，你在学习风格而不是问题。你在学习风格，你在学习问题的社会学。你学到的另一件事是，当你在这些组织做化学家时，你知道，订购一个分子来制造和测试有点痛苦，对吧？所以会发生的是你会做出许多许多你不会追求的选择，而你不会写下那些。哦是的，那样做很愚蠢。就像，是的，那是他们的动机。就像我们的团队就是做不到。他们可以说，我们可以做这个，但那样成本太高。或者实验室团队会说，化学团队会说：“不，我们现在做不了。”或者，你知道，就像我们没有构建块在

But the key here I think is that those chemists also tend to solve similar problems. Like I might have a career here and I have a style but I also am working on a class of problems, right? And so your style is also, you're learning the style instead of the problem. You're learning the style, you're learning the sociology of the problem. And the other thing that you're learning is that when you're a chemist at these organizations, you know, ordering a molecule to be manufactured and tested is kind of a pain, right? So what will happen is you will make many many choices of things you will not pursue, and you do not write those down. Oh yeah, it would be dumb to do that. And it's like, yeah, that's their motivation. Like our team couldn't just make it. Like they could say like, we could do this, but then they would cost too much. Or the lab team will say, the chemistry team will say, "No, we can't make it now." Or, you know, like we don't have the building blocks on

我不认为它会成功。

I don't think it's going to work.

就像有所有这些原因为什么你可能不会做出选择。然后不做那个选择是一个指纹，它会落在确实被制造出来的东西上。然后你知道当你和一些药物化学家交谈时，他们中的一些人，你知道，祝福他们，他们是迷信的人，有些人会说，哦，这是

Like there's all these reasons why you might not make a choice. And then not making that choice is a fingerprint that is going to land on what does get made. And then you know when you talk to some medicinal chemists some of them are, you know, bless them, they're superstitious people and some of them will say like oh this is

是的，我每次都提起这个，你们药物化学家是我见过的最迷信的人之一

yeah I bring this up like every single time you medicinal chemists are some of the most superstitious people I've ever met

对，所以就像，哦，这个构建块在这个项目中起作用了，这个项目进入了临床。我非常喜欢它。也许它有助于溶解性。所以我要把它放在我喜欢的东西里。然后你的模型会学到，哦，化学家喜欢这个，因为他们放了它，所以这意味着它可能有效。所以它学到了所有这些奇怪的东西。所以我们非常害怕这个。所以我们采取了，你知道，我们运行了这个模型来识别化学家是谁。然后我们选择不用任何化学结构或公式来表示分子，而只是用创造者概率的向量。所以现在它是一个 1,800 元素的向量，它有一些是谁制造的概率，你可以使用那个表示，然后训练一个模型来预测 Kemble 的活性，它和使用化学公式一样好，如果你结合它们

right and so and so it's like oh well this building block worked in this in this program that made it to the clinic. I'm really fond of it. Maybe it helps with solability. So I'm going to put it in things that I like. And so then your model will learn like, oh, the chemist likes this because they put so that means it's probably working. And so it learns all this weird stuff. And so like we were very afraid of this. And so we took, you know, we ran this this uh this model to identify who the chemist was. And then we chose to represent the molecules not with any chemical structures or formulas at all, but just a vector of the probability of creator. So now it's this 1,800 element, you know, vector and it's got some probability of who made it and you can use that representation and then train a model to predict activity from Kemble and it does as well as if you use chemical formulas and if you combine them

仅根据社会学。

according to sociology only.

是的，如果你结合它们，它并没有显著更好，所以是的，我们真的认为它学到了很多社会学。它没有学到地面真相。它只是学到了试图开发这些分子的人的选择和指纹。这不是对他们的打击，对吧？我的意思是，那不是他们试图做的。他们试图，你知道，通过制造尽可能少的化学系列来获得两种药物。当然，他们会那样做。但当你用废气训练模型时，你可能会遇到麻烦。

Yes, if you combine them, it doesn't meaningfully better and so yeah, we really think that it's learning a lot of sociology. It's not learning ground truth. It's just learning like the choices and the fingerprints of the people who are trying to develop these molecules. And it's not a knock on them, right? I mean, that's not what they're trying to do. They're trying to, you know, get two drugs by making the fewest number of chemical series possible. Of course, they're going to do it like that. But when you train models on the exhaust, you might be in trouble.

所以，这意味着这不是一个很好的机器学习训练集。

So, this means that this is not a great machine learning training set.

我们不这么认为。这很有趣。

We do not believe so. It's funny.

在我自己的经历中，我也遇到过很多类似的教训。众所周知，像分子量这样的东西——你只需要按分子量排序。这可能是你能拥有的最大的单一预测指标之一。是不是 CAST 16，对吧？

In my own life, I've had many of these similar lessons. It's well known that things like molecular weight will just — all you have to do is sort by molecular weight. That's like one of the single biggest predictors you can possibly have. Like wasn't it CAST 16, right?

CAST 16，它并列第一。

CAST 16 where it was tied for first place.

是的。是的。

Yeah. Yeah.

第二名，非常接近。

Second place like it was really close.

非常接近。

It was really close.

是的。所以所有这些人都拿出他们花哨的机器学习模型，你只需按分子量排序，最重的分子更容易找到结合物。

Yeah. So all these people pulling out their fancy machine learning models, you just sort the molecules by heaviest molecules better find binders.

是的。是的。所以，但这不是化学的内在属性。这是数据集的属性。

Yeah. Yeah. So yeah, but but this is not an intrinsic property to chemistry. This is a property of the data set.

是的。是的。这是关于什么被制造、什么被测试的社会学属性，

Yes. Yes. This is a property of the sociology of what gets made, what gets tested,

以及药物化学家的工作方式。

and like the way medicinal chemists work.

更重的分子更可能结合。

Heavier molecules more likely to bind.

很多时候是因为他们看到了对其他性质的改进。这是部分原因。你可能也有一些部分想谈。

A lot of times it's because they're seeing improvements to affect other properties. That's part of it. And you probably got some parts too you want to talk about.

基本上，你有一个化学家，他们发现了一个命中物，然后开始优化它，他们说：“哦，我假设这是结合位点。所以，让我添加这个官能团来结合这里。”然后他们说：“好吧，那是一个更强的结合物。”然后他们说：“哦，那如果我们添加另一个官能团来结合这个东西呢？”最终发生的是他们不断添加东西，直到它尝起来非常美味。

It's like basically you have a chemist who they find a hit and then they start optimizing it and they're like, "Oh, I have a hypothesis that this is the binding site. So, let me add this functional group to bind here." And then they're like, "Okay, well that that was a stronger binder." And then so they're like, "Oh, well, what if we add this other functional group to bind to this thing?" And what ends up happening is they just keep adding things until it tastes really good.

然后是的，直到它足够好。然后，你知道，然后不是使用组合方法，

And then yeah, until it's sufficiently good. And then, you know, then instead of using combinatorial approach,

比如 Izzian 也说了，然后你还有这些其他性质。你说：“好吧，现在我们有了这个非常好的结合物，但它不溶。所以，让我在分子的另一侧，不在结合口袋的地方，加上一些亲水官能团，让它……你知道，你就不断添加这些东西，然后你就得到了这个大而好的结合物。所以，你只需按分子量排序，这可能是大多数公共数据集中最大的单一预测指标。但这与化学无关。

say Izzian said also then you have these other properties. You're like, "Okay, well now we have this really good binder, but it's not soluble. So, let me throw on, you know, some, you know, um, hydrophilic functional or, uh, some, uh, functional groups on the other side of the molecule, which is not in the binding pocket to like make it, you know, and you just sort of you just keep tacking these things on and next thing you know, you get this big good binder. And so, you just sort by molecular weight and it's like the single biggest predictor that you can have in most public data sets. But that has nothing to do with chemistry.

不，我的意思是，一个很好的直觉方式是，如果某个东西有效，他们就会对它做更多的事情

No, like I mean I think a good way of intuition is like if a thing's working, they're going to do more stuff to it

因为他们喜欢它。

because they like it.

是的。

Yeah.

所以你就对它做更多的事情。它变得更巴洛克，或者你有更复杂昂贵的反应，或者你得到一些奇怪的、难以获得的构建块。

So you just do more stuff to it. It gets more baroque or uh you have more like complicated expensive reactions or you know, you get like weird building blocks that are hard to get.

是的。

Yeah.

你知道，如果你不认为它有效，你就不会做的事情。

You know, things that you wouldn't do if you didn't think it was working.

所以这些都是模型会捕捉到的特征。所以是的，这相当可怕。所以

And so it's all these signatures that the models would just kind of pick up. And so yeah, it's quite terrifying. So

是的，这就像马什么的

yeah, this is this is like the horse something

聪明汉斯。你能讲讲聪明汉斯的故事吗？

clever hunt. Can you uh tell the story about Clever Hunt?

是的。是的。所以，这里上一点历史课，在德国有一个马术训练师。我想是 19 世纪末。我的意思是，我写过这个。我应该记得。有一匹名叫聪明汉斯的马。聪明汉斯似乎能做复杂的数学。所以，你会给马看，比如一块写有数学题的黑板。然后马会用蹄子在地上敲，当正确答案到达时停止。所有人都惊叹马很聪明。然后有人想到一个好主意，去找一个不知道数学题答案的提问者。是的。然后马完全失败了。然后发现马只是在看主人的姿势和微表情。所以当它到达，比如说答案七时，主人在六的时候变得非常兴奋，对吧？让它停止。有一篇由德国心理学家发表的精彩论文，实际上评估了他观察到的马主人的微表情和姿势变化，非常详尽。所以是的，如果你渴望无聊的阅读，你可以去查一下。

Yes. Yes. So, for uh so a little history lesson here is uh there was a um there was a horse trainer uh in Germany. I think it was like the late 1800s. I mean, I wrote about this. I should remember it. And uh there's a horse named Clever Hans. And what Clever Hans seemed to be able to do was complicated math. So, you would show the horse, you know, like a like a a chalkboard that's got a math problem. And then the horse would stamp the horse's foot on the ground uh and then stop when the correct answer were, you know, was was arrived at. And everyone was amazed that the horses are brilliant. And then uh someone had the good idea to go get a questioner who did not know the answer to the math problem. Yeah. And then the horse completely fails. And then it turns out that the horse is just looking at like the posture and the micro expressions mostly of the owner. So that when it got to, you know, let's say the answer seven, then the owner gets like real giddy when it's at six, right? to stop. And there's this like wonderful paper uh put out by by a German psychologist um who actually evaluates the the actual microp expressions and postural changes that he had observed on the horse's owner is quite extensive. So yeah, like if you're dying for a silly read, you can go look it up.

那么类比是，你在读脸，而不是

So then the analogy is that you're reading the face and not the

没错。是的。你不理解数学。模型根本不理解化学。它们只是，你知道，记住创造者的社会学。

That's right. Yes. You do not understand math. The models do not understand chemistry at all. They are just, you know, remembering the sociology of of the creators.

明白了。回到 Hermes 和你的认识严谨性，你知道你试图从所有训练数据中构建有用的东西。是的。所以你发现公共数据集不可信，你有一堆你内部信任的数据，但你并没有真正好的方法，或者但现在你需要理解我们如何准确而谨慎地使用这个来理解在现实世界中做有用的事情。那么你如何弥合这个差距

Got it. going back to Hermes and your uh epistemic rigor and you know your in trying to build something useful out of all the data you trained. Yeah. So you find you know public data sets are untrustworthy and you have a bunch of data which you trust internally but you don't really have a good way of or but now you need to understand how do we accurately and carefully use this to understand to do something useful in the real world. So how do you how did you bridge that gap

让我们思考这个？就像我之前描述的那样，我们做了很多这样的分割。但然后对我们来说下一个明显的事情是，你知道我们在内部数据上训练了模型，我们不太喜欢在外部数据上训练模型。所以也许我们应该尝试在外部数据上预测，但从不训练在外部数据上

for us to think about this? Like we did a bunch of these splits like I kind of described earlier. Uh but then the next thing that seemed very obvious for us to do is like well you know we've trained models on internal data and we kind of don't like to train models on external data. So maybe we should try to predict on external data instead but without ever training on external

从未见过它。所以这确实让我们处于一个有利位置，因为我没有看到任何其他团队专门在内部生成的数据上训练模型来完成这个任务。我认为去获取这些其他来源非常诱人，特别是因为外面有那么多可能的蛋白质靶点，因为人们通常不像我们那样筛选那么多。所以我们取了大量的外部分子集合，然后我们预测它们，看看我们在许多不同的蛋白质、许多不同的阳性和阴性样本上表现如何，我们非常仔细地选择了这些。我想说，在 Hermes 论文中，测试总数大约是 1300 万左右，跨越大约 500 个蛋白质靶点。这是一个巨大的数字，然后我们将其与当时最先进的结构模型进行了基准测试，该模型由 Boltz 团队完成，这是一个非常好的模型，我们的模型无法做到那个模型能做的很多事情。但我们能够在这些公共数据集上，在结合问题方面，做到和那些家伙差不多好。当与 Boltz 比较时，需要一些技巧才能进行公平的比较。我认为 Boltz 也在这个数据上训练了。对吗？

without ever seeing it. And so that does put us in a nice position because uh I haven't seen any other group train uh models for this task exclusively on internally generated data. I think it's very tempting to go get these other sources especially because there's so many possible protein targets out there because people don't screen as many as we do typically. And so we took large collections of of external molecules and then we predicted them uh to see how well we would do across you know many different proteins lots of different positive negatives that we chose very carefully. It was uh I want to say it was like the total number of tests in the Hermes paper. I think it's like 13 million or something across like many it's like 500 protein target. It's some huge number and then we can then we benchmarked it against the uh existing what was at the time like the state of the art structural model which is done by the bolts team and so which is a very good model and like ours cannot do that a lot of the things that that model can do. Um but what we were able to do was do about as well as those guys on those public data sets in terms of like this binding problem. when you compared to Bolts, it required some some sort of tricks to to kind of have a fairish comparison. I think Boltz also trained on this data. Was that right?

算是吧。嗯，所以有几个大型集合，我们并不完全清楚他们训练时使用的不同子集。

Sort of. Um, so there are a couple large collections that um we we weren't totally clear on like the different subsets that they had trained on.

我能想到一个特别清楚的例子，我稍后会讲。但我们用到的很多评估，Boltz 团队也做过——只是我们放了更多分子进去看效果。另外，如果你看 Hermes 和 Boltz 的不同表现，我们在某些蛋白质和蛋白质家族上更好，而 Boltz 更差，反之亦然。我们怀疑这是因为训练集中这些的分布不同，所以他们理解得更多一点——或者老实说，可能记忆得更多一点——并且能稍微扩展一下。有一件事对我们来说非常能验证：我们的一个测试集是另一个正交化学库，我们的模型没见过，我们内部用过，但它是文献中已有化学库的副本。所以有一个特别的目标——它是一种叫 BRD4 的蛋白质——我们筛选中的一个相同分子也在 PDB（蛋白质数据库）训练集中，Boltz 几乎肯定见过。我们的模型在那个分子上表现不太好。那是一个遥远的化学库。它们不太擅长。而 Boltz 完美。我们怀疑这是因为 Boltz 已经在训练集中作为晶体结构的一部分见过这个蛋白质和这个单独分子。所以他们很棒——就像，哦对，这说得通。感觉不错。

I could think of one example in particular where it was very, very clear, and I'll talk about that in a minute. But a lot of the evals we used were also serviced by the Boltz team—only we put a lot more molecules in there to see how it would go. Also, if you look at the different performances between Hermes and Boltz, we're better on some proteins and protein families that Boltz is worse at, and vice versa. We suspect this is because of different distributions of those in the training sets, so they understand a little bit more—or, honestly, maybe memorize a little bit more—and they're able to extend that a little bit. One thing that was very validating for us is that one of our test sets was another orthogonal chemical library that our models had not seen, which we had used internally, but it was a copy of a chemical library that had been out in the literature. And so there was one target in particular—it's a protein called BRD4—where there's an identical molecule from our screen that was also in the PDB, the Protein Data Bank training set, that Boltz had almost certainly seen. Our models did kind of bad on that molecule. It was a distant chemical library. They weren't very good at it. And Boltz was perfect at it. And we suspect that's because Boltz had seen this protein and this individual molecule in a crystal structure already as part of the training set. So they were awesome at it—like, oh yeah, that checks out. Feels pretty good.

所以这是否意味着，在你看来，这验证了你的 Scaling（规模扩张）假设？加入更多蛋白质、更多样化的化学库，现在你基本上可以更广泛地填补这个空间？

So does this mean that, in your opinion, this validates your scaling hypothesis? Add more proteins, more diverse chemical libraries, and now you can basically fill in the space more broadly?

我们希望如此。是的，我们希望如此。我们认为，追求大量多样性，特别是追求有意选择的所谓活性悬崖，是做到这一点的方法。稍微展开一下，这要么在蛋白质侧，要么在小分子侧。你有两个看起来非常非常相似的东西，除了一个微小的变化，而这个微小变化驱动了行为的巨大变化？一个经典的例子，药物化学家会谈论的，是这个想法——它有点神话色彩，不完全是真实的——但这个东西叫做魔法甲基。想法是你有一个分子，然后也许你有这个小刺，它存在或不存在，而这个刺的存在或缺失意味着要么它效果很好，要么它根本不起作用。当你遇到这些情况时，也许你的模型会开始学习，就像，天哪，有一个刺可能真的很有意义，所以也许我需要考虑这些非常小的变化。所以我们经常做的一件事——它已经融入 Dells，因为你有，你知道，三分之二的结构单元可能在大量例子中相似，并且有行为变化——然后在蛋白质侧，我们非常有意地工程化那些蛋白质中的突变来做同样的事情。

We hope so. Yes, we hope so. We think that going after large amounts of diversity and in particular going after intentionally chosen what you might call activity cliffs is the way to do this. To unpack that a little bit, it's either on the protein side or the small molecule side. Do you have two things that look very, very similar except for a tiny change, and that tiny change drives outsized changes in behavior? A real classic example that medicinal chemists will talk about is this idea—and it's kind of mythical, not exactly real—but this thing called a magic methyl. The idea is that you've got a molecule and then maybe you've got this one little thorn that is or is not there, and the presence or absence of that thorn means either it works great or it doesn't work at all. When you have those situations, then maybe your model will start to learn, like, man, having a thorn could be really meaningful, and so maybe I need to think about these very small changes. So one thing that we do a lot—and it's baked into Dells because you've got, you know, two out of three building blocks might be similar across large numbers of examples and have variations of behaviors—and then on the protein side, we very intentionally engineer mutations in those proteins to do the same kind of thing.

我有点困惑，因为如果你有这些结构单元，而你的变化大小是多个原子，对吧？一个结构单元在原子方面有多大？

I'm a little confused here because if you have these building blocks and the size of your variation is multiple atoms, right? How big is a building block in terms of atoms?

对。可能像 6 到 30 个左右，如果我们不忽略氢原子的话。但你可以想象，它们本身是非常相似的结构单元，比如只有一个原子替换。

Right. It might be like six to 30 or something, if we're not ignoring the hydrogens. But you can imagine that they're building blocks that themselves are very similar with like a single atom substitution.

哦，我明白了。好的。所以也许有两个结构单元，一个有刺，一个没有刺，然后你可以在序列中看到这一点。

Oh, I see. Okay. So maybe there's two building blocks that have one with a thorn and one without a thorn, and then you can see that in the sequencing.

没错。没错。也许那个刺只是改变了位置。然后当你思考这个时，你会想，好吧，我可能在目标之间有张力。我可能想要很多更粗糙、更粗粒度的多样性，因为我试图在大量可能非常不同的蛋白质靶标中找到结合物，或者我可能想要非常细粒度的变化，非常非常微妙，以尝试真正教模型这些非常非常微妙的差异。所以当我们考虑设计自己的化学材料时，我们必须跨越这种张力。所以是的，我们经常思考这个问题。

That's right. That's right. And maybe that thorn just changes position. And then as you're thinking about this, you're like, okay, well, I might have tension between my goals. I might want lots of rougher, more coarse diversity because I'm trying to find binders across large numbers of protein targets that may be very different, or I may want super fine-grained changes that are very, very subtle to try to really teach the models these very, very subtle differences. And so as we think about designing our own chemical material, we have to span this kind of tension. So yeah, we think about it a lot.

而这两者都很重要。

And both of those matter.

两者都很重要。

Both of those matter.

在蛋白质侧也很重要，对吧？

And it also matters on the protein side too, right?

在蛋白质侧非常重要。所以当你观察当前的共折叠模型时，几个不同的团队已经展示了这一点。你可以取一个蛋白质，它有一个氨基酸序列，编码某种蛋白质，其中有一个大口袋，也许这个蛋白质在训练集中有代表。所以模型相当清楚当里面有一个分子或小分子化学物质时它看起来是什么样子。所以你有这个口袋，一个进入口袋的东西。你可以改变你给模型考虑的氨基酸序列，你可以用它来破坏口袋。你可以像用水泥填充一样填满口袋，或者你可以完全移除它，这样那里什么都没有，或者它消失了，或者它大得多。你做任何这些事，至少我们目前看到的是，如果你使用当前最先进的模型去看模型会做什么，它会将小分子塞进口袋曾经所在的地方。

It matters very much on the protein side. And so as you look at current co-folding models, a couple different groups have shown this. You can take a protein and it's got this amino acid sequence and it codes some protein that's got a big pocket in it, and maybe that protein is represented in the training set. So the model has a pretty good idea about what it looks like when there's a molecule in there or small molecule chemical in there. So you got this pocket, a thing that goes in the pocket. You can just change the amino acid sequence that you give to the model to consider, and you can use that to break the pocket. You could fill the pocket in like you're filling it with cement, or you could remove it entirely so that there's nothing there or it's gone or it's much bigger. You do any of these things and you get at least what we've seen so far. If you use current state-of-the-art models to go see what the model will do, it will shove the small molecule in where the pocket used to be.

好的。

Okay.

因为它只是记得它在哪里。

Because it just remembers where it is.

是的。

Yeah.

那很糟糕。

And that's bad.

所以这就像看大规模结构的东西，不那么——或者至少假设是它在这样做。

So it's like looking at the large-scale structural things, not so—or like the hypothesis anyway is that it's doing this.

是的。假设是它不太理解，你知道，当你在蛋白质中有一个口袋时，会有少数氨基酸的氢键位置会伸出来。

Yeah. The hypothesis is that it doesn't quite understand, you know, when you have a pocket in a protein, there's going to be a small number of amino acids that will have hydrogen bond positions that are going to be sticking out.

这里面可能有一点 π-π 堆积，还有其他驱动这种化学性质的因素，这些因素本当非常重要。在物理世界里测量它们的时候，它们确实非常重要。所以如果你把它们破坏掉，而模型又理解物理，那它大概应该知道不能硬把一个分子塞进去。问题就在这里——很多这类结构训练集里没有好的负样本，因为能被收录进去的都是……

There's going to be maybe a little bit of pi stacking and these other things that drive this chemistry, and those things really ought to be very, very important. And they are very important in the physical world when you're measuring them. So if you break them and the model understands the physics, then it probably should know that it can't shove a molecule in there. And that's the thing — a lot of these structural training sets, there aren't great negatives, because what makes it into there is...

对，问题就在这儿，对吧？它们没有很多真正能破坏这些行为的变体。

Yeah, that's the thing, right? They don't have a lot of these variations that really disrupt these behaviors.

而且没人会拿一堆这种新变体去做晶体结构，就像我们现在正在做的这样。

And no one's going to make crystal structures with a bunch of these new ones, like we're currently doing first.

但你怎么获得负样本呢？

But how do you get negatives?

我们获得负样本的方式是，我们对基线条件下这些分子数量的分布有非常好的理解。

The way that we get negatives is that we have a really good understanding of the distribution of the number of those molecules in a baseline condition.

好。

Okay.

所以当我们做筛选时，我们知道——哦，好，这些结合得很多，因为当我们把蛋白拉出来时，我们正在富集它们。现在我们把口袋破坏掉，它们就不再被富集了。好。它们不再进去了。我们做的那个单一氨基酸替换，让一批这样的化学系列都掉出去了。

So when we do the screen, we know — oh, okay, these bound a lot, because now we're enriching for them when we pull the protein out. And now let's break the pocket, and they are no longer enriched. Okay. They are no longer going. That one amino acid substitution that we made is making a bunch of these chemical series drop out.

但这是一次性实验。这不是 Adele。

But this is a one-off experiment. This is not Adele.

不，这就是 Adele。

No, this is Adele.

哦，不。所以你不是说你破坏了它。你是说，我找到它被破坏的那个案例。

Oh, no. So you're just — you're not saying you broke it. You're saying I find the case where it was broken.

不，是我们破坏它。我们破坏。所以我们拿——假设我们筛选一个蛋白，口袋是完好的。

No, we break it. So we break. So we take the — so let's say we screen a protein and the pocket is intact.

嗯。

Yeah.

听起来不错。好。我们得到一堆命中。

Sounds great. Okay. We get a bunch of hits.

嗯。

Yeah.

嗯，我们再进一步。假设我们知道别人已经鉴定出一个能进入那个口袋的化学分子。

Um, let's go even further here. Let's say that we know someone else has already identified a chemical that goes in that pocket.

我们买来它。把它塞进去。再跑一次筛选。

Let's buy it. Let's stick it in there. Run the screen again.

嗯。

Yeah.

这意味着所有通常进入口袋的化学分子都会被高浓度的那个——哦，你有竞争性实验。

That means all the chemicals that normally go in the pocket will be displaced by a high concentration of the one that — oh, you have a competitive assay.

哦，我们一直做。

Oh, we do all the time.

好。好。

Okay. Okay.

我们一直这么做。好。然后你说：“好，这看起来相当不错。我相信这个。”然后你说：“好，现在我们知道别人已经表征过的这个化学分子进入这个口袋，因为他们可能做了晶体或其他什么。”嗯。很好。我们看看那个口袋里的氨基酸，然后我们改变为我们制造蛋白的 DNA 构建体。我们现在物理上突变这一个残基。所以现在这个特定的氨基酸——现在我们放一个体积大的进去，口袋就填满了。

We do this all the time. Okay. And then you say, "Okay, well that looks pretty good. I believe that." And then you say, "All right, now we know that this chemical that someone else has characterized goes in this pocket, because maybe they made a crystal or some other thing." Yeah. Great. Let's look at the amino acids in that pocket, and then we will change the DNA construct that makes the protein for us. We will now mutate physically this one residue. So now this particular amino acid — now we put a big bulky one in there and the pocket's full.

比如换一个不同的侧链之类的。

Like a different side chain or something.

不同的——这些你都能做。

Different — all these things you can do.

嗯。现在这真的很容易，因为我们有 AlphaFold，还有 Boltz 或所有这些结构模型，你可以用它们来设计你的蛋白，就像你——

Yeah. And this is really easy now because we have AlphaFold and you know Boltz or all these structure models you can use to design your proteins as you —

你可以用这个——你可以用——而对于表征得没那么好的东西，你知道，我们用保守性分析。你知道，我们盯着看——我们有结构生物学家团队，看看这些东西，哦对，你知道，这位 Edward Craft 太厉害了。所以，好，现在我们要做这一个突变。我们再做一次筛选。哦看，当我们用那个堵住洞的东西去打它时被踢出去的所有化学分子——当我们破坏口袋时它们也不见了。所以现在我们有非常好的证据表明它们都进入那个口袋。

You can use this — you can use — and for things that aren't as well characterized, you know, like we use conservation. You know, like we stare at — we have a structural biologist staff and kind of look at these things like, oh yeah, you know, this guy Edward Craft is amazing. And so, all right, so now we're going to make this one mutation. Let's do yet another screen. And oh look, all the chemicals that got booted out when we hit it with that one thing that's blocking the hole — they're also gone when we broke the pocket. So now we have really good evidence that they're all going into that pocket.

对。

Right.

对。然后关键点来了，你知道，蛋白化学家进来说：“如果你做了突变，你怎么知道你不是把整个东西都破坏了？”对吧？合理的问题。你说：“好，我们只失去了被那个竞争物踢出去的化学系列。所有其他在蛋白其他地方结合的都没变。”它们回来了——

Right. And then here's the point where, you know, like the protein chemists come in and they say, "Well, if you made a mutation, how do you know that you didn't just break the whole thing?" Right? Reasonable question. You say, "Okay, well, we only lose the chemical series that got kicked out by that competitor. All the other ones that are binding everywhere else in the protein are intact." And they come back —

完好。嗯。所以你不是把整个蛋白都破坏了。

Intact. Yeah. So you didn't just break the protein.

你不是把整个都破坏了，对吧？然后你想，好，如果你一周要筛选一百个蛋白构建体，我们大周就是这么做的，就像——抱歉，所以你以某种方式破坏了口袋。

You didn't just break it, right? And so, and then you think, okay, well, if you're going to be out screening a hundred protein constructs a week, which is what we do on big weeks, like — sorry, so you've broken the pocket somehow.

哦，嗯。

Oh, yeah.

那你怎么也发现还有其他东西开始连接、结合到蛋白上？

Then how do you also find that there are other things that start to connect to bind to the protein?

所以，当你改变周围的东西时，你可能会以某种方式改变形状，从而鼓励其他东西结合。我们肯定见过——

So, when you change things around, you might change the shape in a way that does encourage other things to bind. We have definitely seen —

那对你们来说是一个信息来源吗？

And is that an information source for you?

是的。是的。所以另一件非常好的事，DEL 的好处是你可以——你可以想象，如果你不是从大数量来思考，哦，我有这一个化学分子在结合，现在我破坏了口袋，现在它不见了。

It is. It is. And so another thing that's really nice about this and the nice part about DELs is you can — you can imagine, if you're not thinking in large numbers, that oh, I've got this one chemical that was binding and now I broke the pocket and now it's gone.

嗯。

Yeah.

但由于 DEL 的组合性质，也许我有五个不同的化学系列。就像数百个分子，还有另一组是同样的数百个分子，只是构建块有一个单一变化。所以你有这些重复的例子，就像，哦，这一个小东西极大地改变了群体行为，而且这是一个大群体。所以就像，哦，我现在真的信服了。我可以给你们看一张图，我们——我不知道你们在播客上放不放图片之类的——

But because of the combinatorial nature of the DELs, maybe I have five different chemical series. It's like hundreds of molecules, and there's another group that's the same hundreds of molecules with a single change in the building block. So you have these repeated examples where it's like, oh, this one tiny thing shifted the population behavior enormously, and it's a big population. And so it's like, oh, I am now really convinced. I can show you guys a plot that we — I don't know if you put pictures on the podcast or whatever —

呃，但它非常明显——就像，哦，这个构建块上的这一个侧基明显在与这一个氨基酸残基相互作用，因为我们有大约五个独立的例子表明群体行为因它而改变。太美了。非常非常清楚。

Uh, but it's very obvious — like, oh, this one side group on this building block is clearly interacting with this one amino acid residue, because we have like five independent examples of population behavior shifting because of it. It's beautiful. It's very, very clear.

嗯。来吧，给我们讲讲这张图。

Yeah. Go ahead, tell us about this plot.

我们把这张图调出来了。嗯，我们调出来的原因是，我们认为蛋白和小分子都要有大量变异非常重要。嗯，我们认为这很重要的一个原因是——你希望能让一个试图学习药物化学的模型真正开始把东西拼起来——你在这个小分子的这一部分和蛋白的这一部分之间有这些氢键相互作用。嗯，我们见过我之前谈到的其他工作，关于你可以——如果你在数字上破坏一个口袋，然后用当前最先进的模型，它们很难理解物理，它们还是会把分子硬塞回去。所以这让我们认为它们是在记忆。于是我们想，绕过这个问题的一个办法是，在这些蛋白的结合区域实验性地工程化突变，然后重新筛选。

We pulled this thing up. Um, and the reason that we pulled it up is we think it's really important for there to be a lot of variation in the proteins as well as the small molecules. Um, and a reason that we think this is — you want to be able to get the, you know, a model trying to learn medicinal chemistry to really start to put together — you're having these hydrogen bond interactions between this part of this small molecule and this part of the protein. Um, we've seen other work that I talked about earlier about how you can — if you digitally break a pocket and you use the current state-of-the-art models, they have a really hard time understanding the physics and they'll just shove the molecule back in that thing anyway. So it makes us suggest that they're memorizing. And so we thought one way to get around this is to experimentally engineer mutations in binding regions of these proteins and then rescreen.

So what that looks like physically: you take the DNA of the wild-type protein that you've already screened, and then you use microbiology techniques to change one or more amino acids in that protein so that the functional part where the small molecules are binding is no longer the same. It's different in some way. We have different methods to try to identify which amino acids are the smart ones to break. Sometimes you break ones that you're confident will do things. Sometimes you don't know. But we can do this at a pretty big scale. So it's not a big problem for us to take a couple of wingers at each one. And so then you run the screen, and now when the pocket is broken, when the molecules go, what do the molecules do? Do they still go in there? Do some of them go in there? Is there something common among the ones that stick around? Is there something common about the ones that get dislodged or pitched out? And then you have lots of different examples of these where you might have three to five-ish amino acids in a given pocket that are driving the binding behaviors. Maybe you try to break all of them independently, and then you redo the screens each time and get a good feel for what part of the small molecule is interacting with the part that you have changed. And then when you do this with a large library, you'll have lots and lots of repeated examples of similar molecules. They're all following the same behavioral trajectories. And that way you get a really good feel for, oh, I've got lots of examples. They're all pretty similar, but the one thing that is super common is this one little bit on the small molecule. So I have a suspicion that that is interacting with the pocket that I just broke.

So what we're looking at on this plot is: all those little dots are small molecules from a chemical library that we have designed and made. Along the axes are how well the small molecules bound. So if it's really down the axis, it's binding well. If it's more towards the origin, it's not really binding at all. And then we've colored a couple different blobs, and those blobs are different chemical series. The chemicals look very similar to each other, but the difference is where we've pointed that kind of chemical-looking structure. And so if you're squinting and you look at these, they're all pretty similar, but there's a couple subtle differences, and that whole population we believe is driven by interactions of those differences to the one mutant that we made.

So in this plot there's on the bottom the wild-type target and then on the horizontal or the vertical axis you have the mutant type target. And so there's this red one and the red one is pointing basically up.

It is binding better.

It is binding better. So that particular one is like the mutation made that particular one bind better, which is speaking to the question that I asked a few minutes ago.

Yep.

And then maybe some of the most of these seem to be above the axis. So most of this mutation caused most of these to bind better.

That's correct. Yeah. There's a second mutation in that pocket that I can also share with you guys, where all of that population, all those blobs, get kicked out entirely and they do not bind. So then the pocket was broken in a meaningful way, and then here the binding is increased.

So basically everything will be along the bottom.

That's correct.

Okay. That's great.

So you can give us post-production.

Yep.

Awesome.

So doing this you can map out both, well, you can treat the activity cliff issue both in compound space and in protein space.

That is correct. And this is particularly useful if you're, let's say, working at kinases, where there's like a hundred kinases which differ only by, in the important binding site, you might differ by one residue at a time or something. So this can be incredibly powerful for certain classes like that, in addition to your own sanity.

Yep. That's right. That's right. Yeah. And we do this, you know, as I mentioned, we do this a lot. Yeah. Many, many hundreds of these types of screens.

One of the things you can do with this is Hermes sees this cross attention. So you can actually interpret where Hermes thinks the binding pocket is. So it's learning the binding pocket through these types of maps.

Yes. So one thing that we've sort of written about is using these attention mapping approaches to have a look at where Hermes is looking in amino acid space. And when we first did this work, we were expecting it to either look at the binding pockets. That was like the first intuition. And then maybe like a more measured intuition was like, well, maybe what we'd hope for it to do is maybe look at all those residues because it probably doesn't really know what's going on yet. And then when we looked, it was not looking at either of those things. It was looking at really short sequences of amino acids that were away from the functional part of the protein and also away from part of the protein that looks similar across many proteins. And so for maybe the non-biology people in the room, the way a lot of molecular evolution has worked is that there's a stretch of a protein that does a pretty good job, and that often gets shuffled among other proteins as well. So that you have this conserved region. It's called a domain, or you know, conserved protein domain, and so that part looks pretty similar across many proteins. It's something that biologists will use to infer protein families and relationships and functions. So if they all have the same kind of piece, they're probably doing something kind of similar. And so we expected the models to look at that, but they don't. They look outside that, and they look at these little stretches that people don't think typically do a whole lot. They're like a little bit different. They're non-conserved. There's a lot more variation in those. And we realized that Hermes was probably coming up with a nickname for the protein. It's not thinking about, oh, this is the functional part. It's like, I don't need to learn that. I can remember, you know, 10 or 12 amino acids that are different here than everyone else, and I can just memorize it. And we thought that was bad because maybe many of the, like if you have variations in the pocket or whatever, but they all have these other parts that are similar, then maybe the variations in the pocket mostly don't matter, but some of them do. Then the model might learn, oh, I can just pay attention to these.

Yeah.

令人担忧的是，模型再次没有真正学到任何物理规律。就像，哦，我可以直接记住这些东西。这很像当你试图用共折叠模型破坏口袋时，它们还是硬把它塞进去。就像，哦，它不知道发生了什么。它只是起了个绰号。

The concern was that the model again was not really learning any physics. It's like, oh, I can just memorize this stuff. It's very much in keeping with when you try to break the pocket with the co-folding models and they shove it in there anyway. It's like, oh, it doesn't know what's going on. It's just made this nickname.

是的。

Yeah.

就像，天哪，那你怎么解决这个问题？然后我们就想到了尝试大规模进行这些突变。所以发生的情况是，如果你有一个看起来几乎完全相同的蛋白质，但有一个微小的差异。然后由于那个微小的差异，行为发生了很大变化。你有点在鼓励模型关注所有氨基酸。而对于非保守区域，它不能这样做，因为如果它对这两个行为不同的东西这样做，它就会做得很差。

It's like, oh man, so how do you fix that? And then that's when we hit upon trying to do these mutations at scale. And so what happens is if you have a protein that looks almost exactly the same, but there's one tiny difference. And then the behaviors change a lot as a consequence of that one tiny difference. You're kind of encouraging the models to look at all the amino acids. And it can't do that for the section that's non-conserved because if it does that with these two things of different behavior, it's going to do a bad job.

是的。

Yeah.

所以我们回过头来，用这些额外的突变来训练模型，就像你们看到的这个口袋上的那个突变，然后回头查看注意力图，现在 Hermes 会更均匀地关注所有这些。所以你基本上是在使用注意力图来找到它记住绰号的情况，然后丰富口袋中的突变，以鼓励它转而关注口袋。

And so we went back and did model, you know, train models with these additional mutations like the one on this pot that you guys are seeing and then, you know, looked back at the attention maps and now Hermes will look at all of these things much more evenly. So you're basically using the attention map to find these cases where it's memorizing nicknames and then enrich the mutations in the pocket to kind of encourage it to pay attention to the pocket instead.

是的。是的。而且这不一定只是针对任何一个基因。这更像是一个普遍问题，就像，哦，我不能再有效地使用这个绰号策略了。

Yep. Yep. And it's not necessarily just to be across any one gene. It's more like the general problem of like, oh, I can't use this nickname strategy effectively anymore.

你想教模型，不，不要那样做。

You want to teach the model that No, don't do that.

你不要那样做。不，不，那太差劲了。不要

You don't do that. No, no, that's that's lame. Don't do

呃，我喜欢这个，因为我们看到，我认为机制可解释性最务实的应用是在生物学和可能其他相关科学中，我认为它本质不同的原因是因为你反过来，你从模型学习生物学，而不是仅仅试图解释它像什么，它在想什么。在这种情况下，也许它实际上是那个正向案例，但我看到这么多机制可解释性被应用于生物学，因为你真的试图理解，这些机器学习模型中没有语言。没有语言可以依靠。所以，除非你检查模型，否则你真的不知道它在想什么。

Uh so I love this because we've seen we're I think that the most pragmatic use of mechan has been in biology and in probably other related sciences where and the reason why I think it's qualitatively different is because you go backwards from you're learning about biology from the model rather than the just trying to interpret what like what is like what is it thinking In this case, maybe it is actually that forward case, but I've just seen so much mechan being applied to biology because you're really trying to understand you don't have the language in these machine learning models. There is no language to lean on. So, you really have no clue what it's thinking unless you're inspecting the model.

嗯，这真的很有趣。所以，是的，我们最近发现非常有效的一件事实际上是转向语言模型。

Well, it's really interesting. So, yes, and one thing that we have found very fruitful in the very recent past is actually transitioning to language models.

好的。

Okay.

我们目前正在做的是，呃，我们使用可读的、可解释的英语来表示小分子和蛋白质及其行为。

And what we are currently doing is um we're using um readable interpretable English to represent the small molecules and the proteins and their behaviors.

哇。好的。

Wow. Okay.

结果证明效果相当不错。

And that turns out to work pretty good.

是的。我的意思是，这不是。所以这是因为我想在你告诉我们这个之前稍微框定一下，因为在单细胞领域实际上有一篇非常有趣的论文叫做 gene PT，其中人们，所以我记不清作者了。来自斯坦福，是一个统计学家，甚至不是真正搞生物的。呃，也许他们现在是，但那个人他们取了基因描述的文本嵌入。是的。

Yeah. I mean this is no. So this is because I want to just frame this a little bit before you tell us about this because this there's actually a really interesting paper in the single cell uh domain called gene PT in which people so I can't remember the authors. are from Stanford and it was somebody who was a statistician not even really in bio. Um maybe they are now but that person they took the textual embeddings of the descriptions of genes. Yep.

然后他们使用那些，所以就像你为 Hermes 使用 ESM 嵌入。是的。

And then they use those and so like you're using the ESM embeddings uh for Hermes. Yeah.

所以你使用的是 ESM 嵌入。他们只是说让我们只使用文本嵌入。这的美妙之处在于，现在你有无数年的人类时间投入到理解每个基因中，而这些都在某种程度上被这些文本嵌入所捕获，即嵌入的知识。所以就像，是的，所以请告诉我们你在做什么。

So you were using the ESM embeddings. They were just like let's just use the textual embeddings. And the beauty of this is because now you have like a gazillion years of human time put into understanding each of these genes and that's all captured to some extent by these textual embeddings that knowledge of the the embedding. So like yeah so please tell us about what you're doing.

是的。所以我们不久前开始了这个项目，因为我们意识到的一件事是，我们有这个不断增长的活性悬崖集合和相同化学物质在大量蛋白质中的概况。我们想知道我们是否真的可以使用推理智能体去探索它，并在实验室中以循环的方式使用它。所以这要求我们开始使用推理模型和语言模型，以便让它们去玩。所以看起来像这样：我们取了一些开源语言模型，比如我们做了一些 quins 和 GLM，你知道其他人正在使用的东西，我们教它们药物化学，方式是你在这里有一个结构，用自然英语描述它，用自然英语给我两个分子之间的差异，然后我们做到了这一点，我们说好的，这里有一些是结合物的分子，这里有一些不是结合物的分子，你能告诉我们为什么你认为结合物在结合吗

Yeah. So we started this project uh a little while ago uh cuz one thing that we realized is that we had this like growing collection of activity cliffs and profiles of the same chemical material across large numbers of proteins. And we wanted to know if we could actually use reasoning agents to go explore it and use it in a lab in a loop kind of situation. And so that demanded that we started working with reasoning models and language models in order to kind of let them go play. And so what that looks like is we took some open source language models like we did some quins and GLM and you know the stuff that other people are using uh and we taught them medicinal chemistry in ways where you say here is a structure describe it in natural English give me the diff between two very between two molecules in natural English and then we got it to the point of where we say okay here are some molecules that are binders here are some that are not binders can you tell us why you think the binders are binding

然后你也可以给它硬负样本，所以你有这两个集合，认为好吧，我认为这个化学系列是这个，你知道，它会用可读的英语写下这些东西，我们认为这很棒。呃，然后我们说，好的，现在你知道了这些基本概念，我们将给你 10 个结合物和 10 个非结合物，这些是封闭的，让你决定为什么你认为结合物在结合，现在我们将给你机会运行你自己的实验。这里有数百万分子的目录。你为什么不挑选 5 或 10 个并测试你关于它们为什么结合的假设？然后模型会进入我们已经收集的数据。它会提取出它认为最有信息量的选择。它会像编码沙盒一样立即得到答案，然后它会更新其关于结合的假设。最终目标是在尽可能少的步骤中获得最佳结合物，非常像药物化学家会做的那样。然后我们只是

and then you can also give it hard negatives so you have these two collections that think okay well I think this chemical series as this, you know, and it will write these things down in in readable English, which we think is great. Uh, and then we said, okay, now that you sort of know these basic concepts, we're going to give you 10 binders and 10 non-binders that are closed, ask you to decide why you think the binders are binding, and now we're going to give you the opportunity to run your own experiments. Here is a catalog of millions of molecules. Why don't you pick five or 10 and test your hypothesis about why they're binding? And so then the model will go into the data that we have already collected. It will pull out what it thinks are the most informative choices. It will get back the answer instantly like it's a coding sandbox and then it will update its hypothesis about the binding. And then the ultimate goal is to get the best binder in as few steps as possible very much like a medicinal chemist would do. And then we just

你这样做是基于你预先运行的实验，但它是在数据库中查找。

you're doing this based on your a priaryy run experiments but it's looking up in a database.

没错。

That's right.

对。所以现在你基本上创建了强化学习展开。这是一个大规模的强化学习展开，一直在增长，充满了这些活性悬崖和不同的蛋白质变体。所以现在我们有，所以你可以使用这个，因为我们知道当你以这些小的活性悬崖方式改变分子时，这是一个因果变化，就像我们知道当你把这个东西放上去时，行为以可重复、可测量的方式改变。所以你也可以使用这些轨迹来提出推理痕迹，规模相当不错。所以现在我们有这些智能体在数据集中运行。我们只给它们 10 个结合物和 10 个非结合物。你给它们 100 个分子的预算，它们会经历，你知道，大概 100 到 2 次调用，它们只是在处理这个长期的多步推理问题，然后在最后得出它们认为最好的答案。

Right. So now so you're basically created reinforcement roll out. It is a reinforcement roll out of of a large scale that's growing all the time that's full of these activity cliffs and full of different protein variations. And so now we have and so you can use this and since we know that when you change the molecules in these small activity cliff ways it is a causal change like we know that when you put this thing on there the behavior changes in this reproducible measurable way. And so you can also use these trajectories to come up with reasoning traces kind of at a pretty good scale. And so right now we have these agents running through the data set. We just give them 10 binders and 10 nominators. You give them a budget of a 100 molecules and they'll go through, you know, probably like a 100 to two calls where they're just going through this long manystep reasoning problem and then come up with what they think are the answers to the best at the end.

那你为什么觉得这比 Hermes 之类的方案更优，或者比把 Hermes 那套方法 Scaling（规模扩张）——就是搞某种注意力机制——更好？

So why do you think this is superior to something like Hermes or maybe like scaling up the Hermes approach where you're building some attention thing?

对，我们这么认为。我们觉得它对记忆问题有更强的抵抗力。虽然我们对 Hermes 的注意力机制表现很满意，但当你把这些突变体都扔进去时，随着模型参数增加，它们也会记住这些突变体，对吧？我们希望这个东西真正理解这些因果变化，而使用自然语言似乎是这个问题的一个很好的正则化器。我们不知道还有别的什么问题，能有一个物理接地、如此可扩展、如此快速的强化学习循环。你可以想象，当你在做实验室在环的事情时，你要去造机器人、攻克这些问题，它们也必然需要有限的资源。它们仍然要在最初决定做什么、不做什么，然后跑这些循环会花更长时间。如果有一种可扩展的方式来测量那个景观，然后不断把它做得越来越大，让智能体在其中玩耍，我们认为对这个特定问题来说，进展可能会快得多。

Yeah, we think so. We think it's much more resistant to the memorization problem. While we are very happy with where the attention goes with Hermes, when you throw in all these mutants, as you add more parameters to a model, they're going to memorize the mutants too, right? We want this thing to really understand these causal changes, and it seems like using natural language is a pretty good regularizer for the problem. We are unaware of other problems where you have a reinforcement learning loop that is physically grounded, that is this scalable, this fast. You can imagine when you're doing lab-in-the-loop stuff, where you're going off and making robots and going after these problems, they will also necessarily require limited resources. They will still make some choices about what to make in the first place, what not to make in the first place, and then doing those loops will take a little bit longer. If there's a scalable way to measure that landscape and then continually make that bigger and bigger and bigger and to allow the agents to play in it, we think it's probably going to go a lot faster for this particular problem.

这很有意思。是的。我是说，我会说它接近了——Zera 的 Excel 正在接近，但他们实际上不是语言；他们编码成序列。所以某些方面扰动是相似的，但这是不同的。然后 Laya 也与此押韵，但

That's interesting. Yeah. I mean, I would say it's close — it's Excel from Zera is getting there, but they're not actually language; they are encoding as a sequence. So in some ways the perturbation is similar, but this is different. And then Laya also rhymes with this but

确实。这些人都非常聪明，在推动这些自动化的事情。

It does. All these guys are really smart and pushing these automation things.

是的。是的。但他们没有做这些 pull 筛选

Yeah. Yeah. But they're not doing these pull screens

然后把它用作——据我所知，他们还没弄清楚如何把它变成单纯的数据库查询。这样你就能做大量的 rollout。

and then using that as — they haven't, as far as I know, haven't figured out how to make that into just a database lookup. So you can do huge number of rollouts.

对。对。

Yes. Yes.

所以现在，我们喜欢这样来思考：如果你是一名药物化学家，在一家制药公司工作，你会做一件叫做构效关系的事情。意思是，哦，当我以这种方式改变分子时，它会改变行为。所以我试图针对这些不同的性质进行优化。你会经历所谓的 SAR 活动。那是早期药物开发。让我们把分子做得更好。在你的职业生涯中，你可能会看到几十个这样的活动，也许 50 个。

So right now the way that we like to think about it is: if you're a medicinal chemist and you're working at a pharmaceutical company, you do this thing called a structure-activity relationship. What that means is, oh, when I change the molecule in this way, it changes the behavior. And so I'm trying to optimize for these different properties. And you go through what's called an SAR campaign. That's early drug development. Let's make the molecule better. You might see a couple dozen of these, maybe 50 of these in your career.

所以现在，你知道，上周末我们提取出了

And so right now, you know, last weekend we pulled out

超过 10,000 个。而我们甚至没有特别努力，对吧？

over 10,000 of them. And we're not even trying that hard, right?

对吧？

Right?

所以是的，现在就是机会。现在我们将被限制在我们已经筛选过的空间里。然后最后，你可以让模型从比如目录中做出额外的选择。然后你可以在未来去订购那些并重新测试它们，就像你可能对传统的实验室在环方法所做的那样。但现在我们给这些智能体 10 个结合物，给它们 100 的预算，它们会在那个预算内从 650 万个中选出最好的结合物。所以看起来它是有效的。

So yeah, and so now is the opportunity. Now we'll be restricted to the space of things that we have screened. And then at the end, you can ask the models to make additional choices from say catalogs. And then you can go, in the future, go order those and retest them like you might for a traditional lab-in-the-loop approach. But right now we give these agents 10 binders, give them a budget of a hundred, and they'll get the best binders out of six and a half million in that budget. And so it seems like it's working.

是的。所以请多讲讲为什么你认为这是——我理解它是个更好的正则化器这个论点。但你也为什么相信这是有效的？

Yeah. So tell us more about why you think this is — I get the argument about it's a better regularizer. But also why do you believe that this is effective?

对。所以你可以想象，嗯，它不会把东西错得太离谱，但也不会把东西弄得太对。但

Right. So you can imagine, well, it doesn't get things wrong so badly, but it also doesn't get things that right. But

这是——我认为是因为你给它机会做出所有那些选择。

it's — I think it's because you give it the opportunity to make all those choices.

我是说，这就像运行这些编码沙盒。你只是有非常快速的强化奖励信号。但问题是，你还必须包括人类不会做出的选择，因为我们正在测量所有蛋白质上的所有分子。

I mean, it's like running these coding sandboxes. You just have really rapid reinforcement reward signal. But the thing is, you also have to include the choices that a human being would not have made, because we're measuring all the molecules across all the proteins.

是的。所以它必须能够做出反直觉的决定，并得到反馈。

Yeah. So it has to be able to make the counterintuitive decision and get feedback on that.

它必须有反事实。它必须犯错误，你需要确保这些错误存在于数据中，或者未来潜在的错误存在于数据中，方式不会真正偏向于让它学习化学家的社会学指纹。

It's got to have the counterfactuals. It's got to make mistakes, and you need to make sure that the mistakes are present in the data, or future potential mistakes are present in the data, in ways that are not really biased for it to learn chemists' sociology fingerprints.

是的，这真的很迷人。所以通过所有的语言预训练语料库，它对一些基础化学有一种直觉，我认为大多数药物化学家可能都有概念。现在你给它一种无限快速的实验室在环。

Yeah, that's really fascinating. So through all of the language pre-training corpus, it kind of has an intuition for some basic chemistry that I think most medicinal chemists might have an idea of. And now you're giving it sort of like an infinitely fast lab in the loop.

没错。抱歉。

That's correct. Sorry.

也许还有推理。我会说推理能力可能更重要，对吧？能够说，哦，在我的强化活动中，我学到了东西，现在这让我能够推理下一个更难的问题

Maybe reasoning too. I would say that the reasoning skills might even be more important, right? The ability to say, oh, in my reinforcement campaign, I've learned things, and now that allows me to reason about the next harder question

然后继续做出选择。现在我们对早期信号非常兴奋。而且，我们是个小作坊，对吧？我们只有 10 个人。我真的很期待当我们的实验语料库以目前速度的 2 倍、10 倍、100 倍增长时，因为那样这些智能体探索的空间量会急剧扩大。

and then just keep making choices. And right now we're really excited about the early signal. And also, we're a tiny shop, right? There's only 10 of us. I'm really looking forward to when our experimental corpus grows at 2x, 10x, 100x the rate that it currently does, because then the amount of space for these agents to explore expands dramatically.

让我们谈谈那个空间，因为我的理解——这是我从和 Claude 交谈以及谷歌等等得来的——但我的理解是 DEL 有限制，它们倾向于产生平面分子或更平的分子。

Let's talk about that space, because my understanding — and this is from me talking to Claude and googling and whatever — but my understanding is that DELs are limited in that they tend to create flat molecules or flatter molecules.

这取决于构建模块。好的。这取决于构建模块。是的。所以你可以想象，你会受到设计的一些限制，比如，我要拿一些数量的构建模块，通常是两到三个，然后以某种方式把它们拼接在一起。所以当我们做化学库设计选择时，通常——我们现在经常跑的一种是分叉设计。它像一个 Y 形分子。然后还有其他的是线性形状的分子。还有其他的像是圆形、大环形状的东西。然后你必须考虑 DNA 会去哪里，那会如何给我带来问题，然后还有构建模块的商业可用性以及拼接它们所需的化学。你可以走定制路线，需要更长时间。对我们来说，更快更有吸引力。是的。肯定会有化学空间的角落我们可能不擅长。然后你必须考虑，哦，未来的库应该有这种几何形状。然后我还得考虑将要抛给它的不同蛋白质。

It depends on the building blocks. Okay. It depends on the building blocks. Yeah. So you can imagine that you will be somewhat constrained by designs in terms of, oh, I'm going to get some number of building blocks, usually two to three, and then stitch them together in some way. And so when we do chemical library design choices, typically it's like one that we run a lot right now is a forked design. It's like a Y-shaped molecule. And then there's other ones that are linear-shaped molecules. There's yet other ones that are kind of like circular, macrocycle-shaped things. And then you have to think about where is the DNA going to go and how is that going to cause me problems, and then also the commercial availability of building blocks and the chemistries required to stitch those together. You can go for custom stuff, takes longer. It's kind of more appealing to us to go faster. Yeah. And there's certainly going to be corners of chemical space that we're probably bad at. And then you have to think about, oh well, future libraries should have this kind of geometry. And then I got to think about the different proteins are going to throw at it.

而且分布会相当不均匀。用这种方式做出来的整个化合物库，对这类靶点可能非常好，但对另一类靶点可能就很糟糕。于是你就得去想：好吧，我到底在优化蛋白质空间的哪些角落？以及这些张力。

And there'll be a pretty uneven distribution. A whole chemical library made this way might be really good for these types of targets, and it might be really crappy for these other types of targets. And so then you get to think about, all right, what corners of protein space am I optimizing for? And those kinds of tensions.

对。所以听起来这个论点是：好吧，我们知道我们并没有优化整个空间，但已经比过去多很多了，所以这是进步。然后随着我们在这方面做得更好、理解更深，或者预算更大之类的，我们就会扩展到那些我们知道有所欠缺的角落——或者也许我们甚至不知道自己欠缺，但我们会发现。

Yeah. So it sounds like the argument is, well, we know that we're not optimizing the whole space, but it's a lot more than it was before. So that's better. And then as we get better at this and understand, or we have a bigger budget or whatever it is, then we'll expand to these corners that we know that we're lacking in — or maybe we don't even know we're lacking in, but we will find that out.

对。正是如此。正是如此。

Right. Exactly. Exactly.

我们要不要进入最后一个环节？

Do we want to move on to the last phase?

好啊，可以。来吧。看看会怎么样。

Yeah, you can. Go for it. Let's see how it goes.

好。给听众一点背景：Ian 要求——我想确认这是他的主意，不是我的——搞一个，就叫它“别问我”或者“放马过来”环节，我们提出我们能想到的最难的问题。

All right. So for the audience context, Ian requested — I want to confirm that this was his idea, not mine — a, call it a "don't ask me" or "come at me, bro" section, where we ask the hardest problems we can come up with.

你们大概也能看出来，从我这儿问出来的问题不会太难。

Which, you can probably tell, is not very hard from me.

对。

Yeah.

我觉得我的难题已经问完了。

I think I've asked my hard questions already.

我们只有大约 24 小时准备，所以我觉得这些问题大多是 Claude 能想到的最难的问题，不过反正，我们看看会怎么样。对，这个环节的背景是：我上周参加了一个癌症与微生物治疗会议，会议的形式是一半时间做内容报告，另一半时间是非常激烈的拷问。那既可怕又鼓舞人心，所以现在我愿意在公开场合也来一次。

We only had about 24 hours to prepare, so I think most of these will be Claude's hardest questions, but anyway, we'll see how this goes. Yeah, the context for this is I was at a cancer and micro therapeutics meeting last week, and the format of the meeting is half of the time is for content presentation and the other half is for very aggressive grilling. And it was horrifying and inspirational, and so here I am willing to do it in public.

来吧。

Let's do this.

正如 R.J. 刚才提到的，DEL（DNA 编码化合物库）往往具有非常特定的性质，你可以很擅长优化某个 indel，但最终它们常常在某些方面失败，而这些方面很难转化为真正的药物。

DELs, as R.J. was just alluding to, often have very specific properties, and you can get very good at optimizing an indel, but ultimately they frequently fail at certain things which are hard to translate to actual medicine.

对。

Yeah.

所以某些失败模式，比如任何容易组装的东西也容易拆解，对 ADMET 性质不利。你能扩展的空间有限，因为你已经用掉了一些组合构建模块，等等。总之，这在长期超出库的泛化方面有什么限制？

So certain failure modes being such as anything which is easy to assemble is easy to disassemble, bad for ADMET properties. You only have so much room to build on because you've already used some of the combinatorial building blocks, and so on. Anyway, how is this limiting in terms of long-term out-of-library generalization?

我觉得很大程度上取决于构建模块的可获得性，以及是否愿意做定制合成。所以，尤其是对于精细映射这类工作，能够说把氮原子沿着环在所有那些位置上都移动一遍，是非常非常理想的，但对你目前生成的各种构建模块来说，商业上不太可能都拿到。所以这会成为失败模式吗？当然会。是的。是的，我很希望能拿到更多这类东西，加强我们的定制合成。这是我们正在考虑的事。但现在，想拿到这些很有挑战性。

I think a lot of it is going to be somewhat dependent on building block availability and one's willingness to go into custom synthesis. So, especially for the fine-grained mapping stuff, being able to say, move a nitrogen all the way around the ring at all of those positions, is something that's very, very desirable, but it's also something that's pretty unlikely that you're going to get commercially for all sorts of building blocks you're currently generating. So is that going to be a failing mode? For sure. Yes. Yes, I would love to be able to get more of those and beef up our custom synthesis. That's something that we're looking at. But right now, trying to get those is challenging.

Claude 可能会想到的一个批评，关于分子容易组装也容易拆解：很多年来，在 DEL 化学之间连接构建模块最常见的方式是酰胺键或肽键。这些在体内相当容易被消化。但在过去几年——其实是过去四年左右——出现了一波非常好的对 DNA 友好的化学方法，在连接过程中不会破坏 DNA，因为这一点非常重要，它们能给你碳碳键、碳氮键这类东西。所以对于日常的简单化学来说其实相当不错。对，所以我其实不太担心这部分。我更担心的是构建模块的多样性，而不是它的来源。

One critique that Claude might be thinking about, in terms of molecules easy to put together and disassemble: for many years, the most common way to attach building blocks between DEL chemistries is amide bonds or peptide bonds. And so those are quite digestible in the body. But in the past few years — really the past four-ish years — there has been a really nice fluorescence of chemistries that are friendly to DNA, that don't destroy it during the attachment process, because that's very important, that will give you carbon-carbon bonds, carbon-nitrogen bonds, these kinds of things. So it's actually pretty good for workaday simple chemistry. Yeah. So I'm actually not too worried about that part. I'm more worried about building block diversity than the sources for that.

那总体上的 ADMET 性质呢？你一直非常专注于结合。

What about ADMET properties in general? You've been focusing very much on binding.

是的。对。就我们做过的项目而言，试着去看合理的 ADMET 和 PKPD 性质——值得一提的是，我们直接从化合物库命中物做到了小鼠体内良好的 PKPD，只用了六个月，而且药物化学循环次数非常少。所以至少我们目前手上的化学物质，或者至少我们做过的那几个系列，在我们手里结果相当不错。我想我们会继续看这种建模效果如何，当然我们也想做更多这类 ADMET 靶点模型，然后思考如何把其他更新兴的模型类型和我们的设计结合起来。

We have. Yeah. So for projects that we've done, trying to look at reasonable ADMET and PKPD properties — for what it's worth, we've gone directly from chemical library hits to good PKPD in a mouse in six months with a very small number of medchem cycles. So at least the chemical matter that we currently have, or at least those few series that we did that for, turned out pretty good in our hands. I guess we will continue to see how that modeling looks, and also of course we want to do more of these ADMET target models and then think about how to incorporate other types of more emergent models along with our designs.

这种速度是由你们创建定制本地组合库的能力驱动的，还是别的什么让你们能在六个月内做到？

Was that speed driven by your ability to create custom local combinatorial libraries, or what allowed you to get there in six months?

我们在最开始设计那些化合物库时，非常针对类药性优化了那个空间，所以我们认为这是部分原因——那是一个非常好的起点。当我们做化合物库设计时，我们有这种类似 Thompson 采样层面的方法，先提出一个优化函数。你说，好，我希望这些库具有这些性质。其中一些性质会是：我想结合这些东西，我想避免这些东西。然后另一些性质会是——哦，还有别的 ADMET 模型方法可以用来预测你的化学物质可能会怎样。于是你把它作为优化函数的一部分。然后你要做的是：好，在给定我的几何结构下我能造出的所有化合物里，我不想造那么多，我想造更少的数量。所以你基本上就是设置并生成一堆合成库，然后用这个优化函数给它们打分。我们对这个库就是这么做的，起始物看起来相当不错。所以我相信我们会继续这么做。

So when we designed those chemical libraries at the beginning, we very much optimized that space for drug-likeness, and so we think that's part of it — that was a really good starting point. When we do our chemical library designs, we have this kind of like Thompson sampling-level approach where we come up with an optimization function. You say, okay, well, I want the libraries to have these properties. And so some of those properties are going to be, I want to bind these things, I want to avoid these things. And then some of these properties are going to be — oh, there's other ADMET model approaches that you can use to make predictions about how your chemical matter might turn out. And so you put that as part of the optimization function. And then what you do is you say, okay, of all the chemicals that I could make given my geometry, I don't want to make that many. I want to make a smaller number. And so you basically set up and generate a bunch of synthetic libraries and then you score them on this optimization function. And so we did that with this library, and the starting material looked pretty good. So I'm sure that we'll continue to do that.

顺着这个思路，你们做的任何东西有没有推进到——就是推进到动物，然后——

Along those same lines, has anything that you've done made it to — so it made it to animals and —

小鼠体内的 PKPD 就是那个。对。

So PKPD in a mouse is that. Yeah.

所以那意味着，你能给小鼠用药，然后看它如何进入小鼠的身体和组织？

So what that means is, can you dose the mouse with the thing and then see how it gets into the mouse's body and tissues?

对。PKPD 就是药代动力学、药效动力学，意思是身体如何让药物在体内移动。

Yeah. So PKPD is pharmacokinetics, pharmacodynamics, meaning how does the body move the drug through the body.

对。然后有没有药物预期活性的生物标志物？所以这里的意思是，这是一种药物，这是我们正在研究的一个用于炎症的化合物。

Yep. And then are there biomarkers of the drug's intended activity? And so what that means here is this is a drug, this is a compound we're working on for inflammation.

于是我们追踪了小鼠血液中的细胞因子水平，它们确实按照我们基于药物靶点的预期发挥了作用。

And so we were following up cytokine levels in mouse blood, and they did what we wanted them to do and expected them to do based on the targets of the drug.

所以，Sean 喜欢问什么时候能赚钱。我想问的是，什么时候人们——

So, Sean likes to ask when money. I want to ask when people—

是的。我的意思是，对于那些我们谈过的人，我们有几个这样的项目，正在和一些人讨论合作，成立新的实体去推进它们。

Yeah. I mean, for those we talked to, we have a couple of these things that we're talking to folks about partnering up to spin up new entities to go chase them down.

好的。

Okay.

所以，是的，当然，所有这些事情都需要一些时间。你得推进这些传统流程——一旦你宣布了所谓的开发候选药物，你还得做一堆关于生产制造和额外安全性方面的工作，才能进入人体试验。所以这些事情都需要更长一点的时间。

And so, yeah, and of course, like all these things take some time. You've got to do these traditional things forward—once you declare what's called a development candidate, then you've got to do a bunch of other work about manufacturing and additional safety work before you can get into people. So all these things take a little longer.

是的。所以你们打算拆分出去，想法是拆分出一个——所以也许甚至内部的人会说，我真的很想把这个东西推向市场，或者试着把它推进到下一步。

Yeah. And so you're going to spin out the idea is to spin out a—so people maybe even internal, they will say like, I really want to take this thing to market or try to get it, like take it to the next step.

是的。是的，我的意思是，我们思考这件事的一种方式是，你知道，我们打造了这个非常出色的引擎，用于数据生成和模型训练，现在正如我们讨论过的，还有这些强化学习推理方法。在大多数其他领域，当你构建一个引擎去做这些事情、去帮助解决这些问题时，你在这个过程中生成的子集本身并不具有内在价值。没有人想卖关于政治理论的推理轨迹，

Yeah. Yeah, I mean, one way that we kind of think about it is, you know, we built this really great engine for data generation and model training, and now as we've talked about, these reinforcement learning reasoning approaches. And in most other domains, when you build an engine to go do that stuff, to go help on those problems, the subset that you're generating along the way is not itself intrinsically valuable. No one wants to sell reasoning traces about political theory,

但就像我们在做所有这些筛选的过程中，会生成一些小小的金子，

but like there are going to be little bits of gold that we have generated as a consequence of doing all the screening

因为你实际上不仅在推理这个模型很好——抱歉，这个分子很好——而且在推理这里为什么它好。这里是为什么它好，而这里是对我们专有的化学材料，它在我们认为别人可能难以成药的蛋白质靶点类别中具有活性，对吧？所以你有了起始材料——你知道，现在我们有的起始材料，在我看来对大约 700 个靶点都相当有选择性。这是个巨大的数字。你不想把它们全部推进，但你知道，作为机会，它们都是具有不同程度经济价值捕获和可取性的机会。所以你想，好吧，我们有了所有这些材料。让我们看看能从中得到什么。让我们看看能从中得到什么。药物发现只占开发药物总成本的一小部分，你可以争辩说，好吧——

because you're actually reasoning about not only that this model is good—or sorry, this molecule is good—but that this here is why that's good. Here is why that's good, and here is chemical material that is proprietary to us that has, you know, activities that are in protein target classes that others might find hard to drug, right? And so you've got starting material for—you know, right now we have starting material that looks pretty selective to us for, I think, like 700 targets. Like it's an enormous number. You don't want to chase all those down, but you know, as opportunities, they are all opportunities with different levels of economic value capture and desirability. And so you think, all right, well, we got all this stuff. Like, let's see what we can get out. Let's see what we can get out of it. Discovery is like a small percentage of the total cost of developing a drug, and you can argue that, well—

你有很多——任何一次发现也许很便宜，但由于失败，你有很多次发现。所以但即便如此，你知道——

you have a lot—any given discovery is cheap maybe, but because of the failures you have a lot of them. So but even then, you know—

要得到一款药，一款药，就像单个候选药物的所有步骤

to get one drug, one drug, like all the steps of a single drug candidate

是的

yeah

一个成功的候选药物，你知道，低的话大概 2 亿美元，高的话 10 亿美元，嗯，取决于你要在谁身上测试以及很多其他因素。但然后你知道，发现阶段只是几百万美元。所以，你在这里解决的是正确的问题吗？

that is successful is maybe, you know, on the low side 200 million, on the high side a billion, um, depending on who you have to test on and lots of other factors. But and then you know the discovery is a couple million. So so are you tackling the right problem here?

嗯，所以我认为有一些早期发现的问题值那么多钱，你知道，合理的金额。那这意味着什么呢？现在有一堆靶点，有非常好的遗传学证据，但用小分子非常难以成药。

Um, and so I think there are a number of early discovery problems that are worth that much, you know, like reasonable amounts. And so what does that mean? There's a bunch of targets right now with like really good genetic evidence that are very challenging to drug with small molecules.

嗯，你知道，一个很直接的例子是，呃，对于很多这些，嗯，呃，细胞因子炎症适应症，人们用抗体来治疗，而去追求一种小分子疗法，你可以作为药片服用，而不是输液，一种不需要，呃，像冷链，你知道，运输，呃，就能到达不同患者群体的疗法。拥有这些东西是非常可取的，

Um, you know, a pretty straightforward example is, uh, for a lot of these, um, uh, kind of cytokine inflammatory indications, folks are using antibodies to drug them and going after a small molecule therapeutic that you could take as a pill instead of an infusion, one that doesn't require, uh, like cold chain, you know, transport, uh, to get to different patient populations. Like having those things is enormously desirable,

意味着你得有冷藏卡车，

meaning you have to have refrigerated trucks,

对吧？是的。就像如果你想把这些——如果你想把这些基于抗体的药物送到其中一些人手中，这就像一个非常困难的问题。

right? Yeah. Like and like if you want to get these—if you want to get these antibody-based drugs to some of these folks, like this is like a very hard problem.

是的。而且非常昂贵

Yeah. And very expensive

而且非常昂贵。所以所以人们真的非常兴奋，你能不能找到一种小分子疗法，做同样的事情，呃，那种针对这些的抗体——

and very expensive. And so and so folks are like really quite excited about, can you find a small molecule therapeutic that does the same thing, uh, that antibody against these—

所以有一些抗体，它——它有效。

so there's some antibody, it's—it works.

是的。所以我们知道它有效。

Yep. And so we know it works.

然后然后你就想，好吧，为什么我们不能用一个小分子去针对同一个靶点呢？

and and so you're like, well, why can't we just go after the same target with a small molecule?

事实证明，由于各种原因，这些蛋白质非常难以成药，但我们倾向于认为，你知道，我们正在开发的技术可以得到好的材料。所以我们确实对很多这些细胞因子有选择性材料。所以这是一个例子，一个非常可取的靶点，当然，要把它推进到真正上市，会有更多的，呃，资本需求，但兴趣足够高。

And it turns out that those proteins are very hard to drug for a variety of reasons, but we like to think that, you know, the technology that we're developing can get good material. And so we do in fact have selective material for a lot of these cytokines. And so that's one example of something that's like a very desirable target where, of course, like there's going to be much more, uh, capital requirements to like get it across the line to actually get it to market, but the interest is sufficiently high enough.

所以基本上你们要毁掉抗体市场了。我的意思是，如果你能的话，那么它对某些东西是好的，对某些东西是很好的，而且你知道，有很多精力被投入到那些团队，因为我们擅长蛋白质折叠，对吧？所以他们正在取得很多进展。

So you're going to ruin the antibody market basically. I mean, if like if you can, like then it's good for some stuff, like it's great for some stuff, and like there's a lot of energy being thrown, you know, at those groups because we're good at protein folding, right? And so like they're making a lot of progress.

也许更广泛地说，我会说，要解决正确的问题。这有点像一个路灯问题，在生命科学中，我们能做足够多可扩展测量的地方是这些早期问题。我们没法真的做一百万次小鼠实验，对吧？就像我们没法做几百万次一期临床试验。所以我们会在这方面变得更好。我们会开始用更好的数据、更好的推理和物理模型，去攀登这些不同的生物抽象层次。但就像现在，我们真正需要做足够多测量的技术，我认为会是这些早期方法。

And maybe more broadly, I would say is like going after the right problem. It is a little bit of a lamppost problem, where the things where we can make enough scalable measurements in the life sciences are these early problems. We can't really run a million mice experiments, right? Like we can't run many millions of phase one trials. And so we're going to get better at that stuff. We're going to start to like go up these different layers of biological abstraction with like better data and better reasoning and physics models. But like right now, like the technologies that we have to really make the number of measurements that we need are going to be like these early approaches, I think.

是的。

Yeah.

是的。我还会争辩说，如果你有一个预言机，某个 AI 预言机说，哦，这个，所有 ADMET 属性都会很好，而且这正好泛化到这个人群。所以你有这个预言机过滤器，那么你在临床试验中的所有失败都会消失。

Yeah. And I would argue also that you can, if you have an oracle, some AI oracle that says, oh, this, like all the ADMET properties will be good, and this generalizes to exactly this population. So you have this oracle filter, then all of your failures in clinical trials go away.

是的，你可以想象一个世界，你可以提出一个——或者你有一些新靶点，你知道，一些新证据覆盖了，呃，然后预言机可以立即给你小分子材料，你会觉得可以直接拿去跑小鼠研究。

Yeah, you can imagine a world where you can come up with an—or you have some new target that you know, some new evidence that covers, uh, and then the Oracle could give you small molecule material instantly that you would feel comfortable running straight to like a mouse study.

对。对。对。所以，但那需要在来自小鼠研究、临床试验等的数据上训练。

Right. Right. Right. So, but so that requires training on data from mouse studies, clinical trials, whatever.

我的意思是，也许，我的意思是，我可以肯定地说，如果你对靶点有信心，ADMET 属性，你需要确保它不会命中所有其他这些脱靶靶点。

I mean, maybe, I mean, I could say definitely like ADMET properties if you're confident in the target and you need to make sure that it's not hitting all these other off targets.

好，我们现在来聊大家最喜欢的那个老生常谈的问题。我最喜欢的老生常谈，就是我称之为机器学习与发现中的“科塞里悖论”。

All right. Well, we're on everyone's favorite pet question. My favorite pet question is what I call the Koseri paradox of machine learning and discovery.

对。对。

Yes. Yes.

好。那么，Shri Kosari，那条推文很棒。我很喜欢那条推文。他几周前上过我们的节目。希望我们能在接下来几周内发布他那期。

Okay. So, Shri Kosari, great tweet. I love this tweet. He was on with us a few weeks ago. We hopefully will release his episode in the next few weeks.

可惜的是，我在那期节目里居然都没提到这个。我太失望了。

Sadly, I actually didn't even bring this up on the episode. I'm so disappointed.

总之，那条推文大致是说：机器学习做药物发现的商业模式到底是什么？因为要构建一个好的机器学习模型，你需要大量数据，需要规模化、高质量的数据。你显然有这些数据。但如果你需要这些数据来训练模型，那你还要模型干什么？因为你直接去收集就行了，而你显然能把这些东西收集得相当好。所以，我们聊了很多数据收集，也聊了很多机器学习。我其实有点好奇：实际的机器学习——它对你们的商业模式是必需的吗？

Anyway, but yeah, the tweet is basically: what is the business model in machine learning for drug discovery? Because in order to build a good machine learning model, you need lots of data and good high quality data at scale. You clearly have that. But then if you need that to train the model, like what do you need the model for? Because you just go collect the stuff and you clearly can collect this stuff quite well. So yeah, we talked a lot about data collection. We talked a lot about machine learning. And I'm actually kind of curious in some ways: like where does the actual machine learning—is it necessary for your business model?

这当然取决于商业模式。我认为，如果你唯一的理由就是去为某个特定靶点拿到好的化学材料，那科塞里当然说得对，对吧？如果你有足够多的化学材料，广泛覆盖所有这些其他靶点，能够驱动对化学行为更深入的机器理解，那你就可能比优化那些早期材料的方式快得多——尤其是如果你能像我们这样规模化地做，尤其是如果你能像我们这样以很低的资本投入来做。我们在 Leash 并没有融很多钱，却筛了海量的东西。再往后退一步看，我提到语言模型这些东西的原因之一，是我们开始怀疑，它是否能带来比仅仅识别新化学材料大得多的能力范围。比如，如果我们真的擅长化学推理，那我们就能擅长很多不同的行业及其应用，而不必亲自去那些不同领域里做筛选。

So I mean of course it would depend on the business model. I think if your only rationale is to go off and get good chemical material for a particular target, then of course Shri is right, right? You could get started if you had enough chemical material broadly cast across all these other targets in a way that can drive a deeper machine understanding of the chemical behaviors than the way by which you optimize those early materials—have the potential to go much faster, especially if you can do it at the scale that we've done it, especially if you can do it at the low capital infusions that we've been able to do. We have not raised a lot of money at Leash, and we've screened an enormous amount. And then I think, getting even stepping back, one of the reasons that I brought up this language model stuff is because we are starting to wonder if it would enable a much larger scope of capabilities than just identifying novel chemical material. Like if we get really good at chemical reasoning, then we can get really good at lots of different industries and their applications without having to physically go and screen in all those different domains.

能举个例子吗？

Like what's an example of that?

假设我们真的把这些预测做得很好了，那我们就能立刻跑去农业领域干活，对吧？你可以去做兽医科学，可以去做材料科学，而且你不必——如果你真的很擅长，你也不必去筛那些蛋白质，因为你对这些东西有了很好的基础理解。同样的道理也适用于所有蛋白质折叠公司，对吧？现在这些模型变强了，就有团队去攻克那些尚未被筛选、或者极难筛选的蛋白质靶点，并拿到非常好的起始材料。比如 novel 那帮人就是这么做的，针对细胞上的一个受体，上面有一小段突变的癌症基因，非常难搞，他们还是做成了。所以你可以想象，也许存在一个世界，我们用小分子材料也能做到这一点。所以我不认为科塞里错了——如果你只是想给这一个东西做药，筛一大堆，然后你已经拿到了那个东西，你就不需要模型了。但这和泛化到所有东西并不完全是一回事。我们希望，如果你有足够多这种工程化的活性悬崖的变异，你实际上就能达到一个可以做所有事情的状态，到那时你就再也不需要把一切都筛一遍了。

So let's say we get really good at making these predictions, then we can run off and go work in agriculture like immediately, right? You could go work in veterinary science, you could go work in material science, and you don't have to—if you're really good at it, you don't have to screen all those proteins too, because you have a good fundamental understanding of this. And the same thing could be argued for all the protein folding companies, right? It's like now that these models are good, there are teams that are going after protein targets that have not been screened yet or are very challenging to screen and getting really good starting material. Like the novel guys did it with this basically a receptor on cells that has like a little tiny piece of a mutated cancer gene that's really hard to do, and they managed to pull it off. And so you can imagine, oh, there could be a world where we do this with small molecule material as well. And so I don't think Shri is wrong in terms of finding like start—you know, if you're just trying to drug this one thing and you screen a lot and then you already have the thing, you don't need a model for the thing. But that's not exactly the same as generalizing across all things. And we are hopeful that if you have enough of this variation of engineered activity cliffs, you will actually be at a point where you can do all things, in which case at which point you don't need to screen it all ever again.

对。而且可以说你们——我不太清楚你们具体融了多少钱。所以，我是说，你们做一轮 campaign，你说的那相当于几轮 campaign，成本是一样的。

Yeah. Well, and arguably you guys—if I don't know exactly how much you raised. So I mean, you're for a campaign, you're talking that's like a couple campaigns that would cost the same amount.

对。对。也许可以给点真实的参照：我们开发并筛选了大约 2,000 种不同的蛋白质构建体，公司历史上总共大概做了 5,000 次筛选，每次至少覆盖 650 万个分子。

Yeah. Yeah. I mean maybe to put some real perspective on it: so we have developed and screened about 2,000 different protein constructs, a total of probably about 5,000 screens in the history of the company, across at least 6 and a half million molecules each.

我不想陷入纯粹的数字游戏，但这肯定几乎和地球上任何其他人都一样多。也许比所有人都多。大概不是。可能有几个团队——如果你了解的话。但就突变数量而言，我觉得很可能比所有人都多。这在我看来是很有可能的。

Which I don't want to get into like a huge numbers game, but it's certainly almost as much as anyone else on Earth. Maybe more than everyone else. Probably not. There's probably a couple groups that—like if you know. But like the number of mutations, I think it's probably more than everyone else. That seems likely to me.

所以，对，我们投入了巨大的精力，让这件事变得非常便宜、非常可扩展。如果我一次只做一个——

And so yeah, like we just put a huge amount of work into making this very cheap and very scalable. And if I was doing one at a time—

对。如果你去找一家 CRO，一次做一个，每个大概要花你 10 万美元，取决于对方给你多少手把手的支持，对吧？

Yeah. If you went to a CRO and you did one at a time, it would probably cost you 100K each depending on how much handholding you got, right?

对。

Yeah.

你刚说五百六十万，是多少来着？650 万个分子。这不算很多分子，尤其考虑到这是组合空间。你知道，大家都说 10 的 40 次方之类的。我是说，很大的数字。显然，你不需要理解整个化学空间。你只需要持续、反复地做出药物。那是你的目标。但这点化学空间真的足够让你在你们的化学空间之外有意义地做点什么吗？

You just said five million six, what is it? 6 and a half million molecules. That's not a lot of molecules, especially since it's combinatorial space. You know, everyone says 10 to the 40 or something. I mean, big numbers. Obviously, you don't need to understand all the chemical space. You just need to make drugs consistently, repeatedly. That's your goal. But is that actually enough chemical space to meaningfully do anything outside of your chemical space?

好问题。令人愉快的答案是——我之前说过，我们只用内部数据训练模型，而且我们能够扩展，并泛化预测到 ChemBL，那是化学空间中非常不同的角落。所以，是的，我们认为这够了吗？远远不够。因此我们前进路线图的一部分，就是在化学空间的不同角落构建更多数量的化合物库，并覆盖我之前谈到的范围：有些是非常细粒度的变化，以便更好地把握活性悬崖；另一边则是大的粗粒度变化，试图捕捉新的蛋白质和蛋白质形状。

Great question. So the pleasant answer is—I said earlier that we have trained models on internal data only, and that we're able to extend and make generalized predictions into ChemBL, which is very different corners of chemical space. So yes, do we think it's enough? Not remotely. And so part of our road map forward is making larger numbers of chemical libraries in different corners of chemical space, and kind of spanning what I talked about earlier, where some are these very fine-grain changes to more appreciate activity cliffs, and then the other side it's these big coarse changes to try to capture new proteins and protein shapes.

抱歉，这不是一个刁难的问题，我只是真的好奇：你们能有意义地预测效力、相对效力吗？比如你们能做 KD 曲线吗，就是滴定越来越大的化合物库？人们在这方面有过一些摸索，试图直接从 DEL 里得到 KD，比如有人会做改变蛋白质浓度之类的事情。

This is sorry, not a hard question, just I'm genuinely curious: are you able to predict like potency, relative potency meaningfully? Like can you do KD curves, where you titrate like larger and larger libraries? So people have wrestled with this a little bit in terms of trying to get the KDs directly out of DELs, doing things like maybe varying protein concentration is something that folks do.

我们看到的一件事是，当我们重复筛选时，我们会得到相同的命中，顺序大致相同，这让我们有点乐观，认为你可以得到这类东西。我们开始更多地转向 SPR、生物物理学相关的测量，试图弄清楚我们是否能想出一个效力阶梯，然后我们可以将其作为我们正在筛选的化学材料的一部分，以试图捕捉这些东西。但我不会说我们已经擅长这个了。

One thing that we see is when we do replicates of screens, we get back the same hits in roughly the same order, which makes us kind of optimistic that you could get these kinds of things. We started to move more into SPR, biophysics-related measurements, try to figure out if we can maybe come up with a potency ladder that we could then be as part of the chemical material that we're screening to try to capture those things. But I won't say that we're good at it yet.

就像你不能用读数作为 KD 的代理之类的。

Like you can't use reads as a proxy for KD or something.

我认为相对排名可以，但我认为读数用于 KD 速率，读数用于 KD 不太行，然后你开始遇到奇怪的东西。所以就像我们遇到过很多蛋白质，结果发现 DNA 条形码本身，有些蛋白质没有被描述为 DNA 结合蛋白，但我们发现它们是，所以你可以想象这可能会影响事物的 KD，你必须弄清楚如何缓解这些问题和那种东西。所以就像是的，我认为相对排名是相当可行的。我认为将其转化为 KD 是我们正在积极努力的事情，我们认为我们可能能够获得一些合理的、有点粗糙的转化，但我不会说我们已经做到了。

I think the relative ranking yes, but I think reads for KD rates, reads for KD not really, and then you start running into weird stuff. So like there's lots of proteins that we've encountered where it turns out that the DNA barcode itself, there are proteins that are not described as DNA binders that we have discovered are, and so you can imagine that that might affect the KDs of things and you have to figure out how to mitigate those problems and that kind of stuff. So like yeah, I think the relative ranking is something that's quite doable. I think translating it into KDs is something that we're actively working on and we think we might be able to get like some reasonable, kind of like sloppy translation, but I wouldn't say that we're there yet.

好的，这个，这个相当残酷。苦涩的教训。是的，我就直接读这个。这个是纯 Claude 的。我在这上面没有想法。苦涩的教训真的是定律。什么阻止了一个资金更充足的实验室在规模上超过你的 Dell 引擎？

Okay, this one, this one was pretty brutal. The bitter lesson. Yeah, I'm just going to read this. This one is pure Claude. I had no thought in this. The bitter lesson is really the law. What stops a better funded lab from outscaling your Dell engine?

哦是的，当然人们会问我们这个。所以原因，我真正想说的是，在 Leash 和 Recursion，我的团队犯了很多错误，意外地撞上了那些只是做实验室工作后果的作弊模式。有很多这样的，所以你必须追查这些，很多与用元数据跟踪批次效应有关，例如，然后你必须在你的组织中建立一种文化，就像好吧，我们必须跟踪所有这些事情，我们必须构建工具跟踪所有这些事情，我们必须使这些工具无摩擦，让做工作的人实际上会记录这些事情，然后你必须回去查看所有数据，回到公司的创立历史，以确保你得到真正一致的结果，然后当你看到大的变化时，你必须确定是什么驱动了它们，然后你必须缓解这些，这就像，就像大量的工作。我们还从头重建了选择过程。我们从头重建了很多蛋白质表达过程。然后我们有所有这些其他代理，我们用来评估蛋白质生产进行得如何，是好是坏，以及这些其他过滤器。这只是需要一种独特的文化、耐心和愿意犯这些错误，然后让团队真正深入思考如何解决它们。我认为除非你有真正对这些实验错误有嗅觉的人，否则复制起来会相当，需要一些时间。

Oh yeah, of course people ask us this. So and the reason, here's what I would really say is that both at Leash and at Recursion, my team has made so many mistakes in accidentally bumping into the cheating modes that just kind of come out of the consequence of doing laboratory work. There are so many of these and so like you have to chase these down and a lot of it has to do with tracking batch effects with metadata for example and so then you have to build a culture at your organization where it's like all right we have to track all these things we have to build tools track all these things we have to make those tools frictionless that the people doing the work actually will record these things and then you have to go back and look at all the data back into the founding history of the company to make sure you're getting really consistent results and then when you see big changes you have to identify what drove them then you have to mitigate those and it's just like it's like a huge huge amount of work. And we also rebuilt kind of from scratch the selection process. We rebuilt from scratch a lot of the protein expression process. And then we have all these other proxies that we use to evaluate how well the protein production went and if it's good or if it's bad and these other filters. And it's just it takes kind of a unique culture and patience and willing to make these mistakes and then have the team like really think very deeply about how to address them. And I think unless you've got folks who really have a nose for these kind of experimental mistakes, it's going to be pretty, it's going to take some time to replicate.

过程模式。

Process mode.

是的，我真的这么认为。有太多事情可能出问题，你必须一直足够注意，在它们出问题时识别出来，它们可能是你可以编写自动过滤器的事情，也许早期不是。所以就像你知道的，一个很好的例子，我很乐意谈论的是，人们经常问我们，我们怎么知道我们的蛋白质是好的？我们有很多方法可以弄清楚这一点。我现在最喜欢的一个是，我们注意到当蛋白质错误折叠时，我们能够在非常粗略的情况下识别错误折叠，比如如果你把蛋白质跑在凝胶上，这是一种蛋白质技巧，你试图弄清楚它有多大，如果它错误折叠，那么你可以检测到，但如果它稍微错误折叠，可能很难。有些药物的作用是重新折叠错误折叠的蛋白质，称为分子伴侣。

Yeah, I really think so. There's just so many things where it can go sideways and you have to be paying enough attention all the time to identify them when they're going sideways and they might be things that you can write automated filters for and maybe not early on. And so like you know, like one pretty good example I'm happy to talk about is people ask us a lot, how do we know our proteins are any good? We have lots of ways to figure this out. I think of one of my favorite ones right now is we have noticed that when proteins misfold and we're able to identify misfolding in very crude cases like if you put the protein run on a gel which is this protein trick where you try to figure out how big it is and it misfolds then you can kind of detect that but if it misfolds a little bit it could be hard. There are drugs whose job it is to refold misfolded proteins called molecular chaperones.

好的。

Okay.

你也可以叫它们校正剂。我们会这么叫它们。

You can call them correctors too. That's what we would call them.

是的。

Yeah.

有一个非常著名的用于治疗囊性纤维化患者的药物叫 lumacaftor。我们的化学库中有很多看起来像 lumacaftor 的化学物质。这个东西的作用是重新折叠错误折叠的蛋白质。猜猜当我们筛选它们时，它会命中什么样的蛋白质？

There's a very famous one that treats cystic fibrosis patients called lumacaftor. We have a bunch of chemicals in our chemical library that looks like lumacaftor. This thing whose job it is to refold misfolded proteins. Guess what kind of proteins it hits when we screen them?

囊性纤维化。

Cystic fibrosis.

那里面的错误折叠蛋白质。

Misfolded proteins in there.

这就是它的用途。

That's what it's for.

所以就像，它就在我们的库里。所以你知道，我们意外发现了这个，就像我们看到这些化学系列不断出现在所有这些不同的蛋白质类型中，就像天哪，那是什么，是 captor 吗，然后你开始看蛋白质表达，然后它进行得如何，就像哦，当事情不顺利时它就会出现，所以现在我们可以用它作为一个过滤器，如果我们认为，如果我们得到一堆 lumac 头，那蛋白质可能没有正确折叠。我们可能不想把它用作训练数据。我们可能想用稍微不同的条件重新运行它，让它对蛋白质更友好。然后我们会看到那些命中是什么样子。所以除非你付出那种注意力，否则真的，我的意思是这真的很难。是的。你必须构建工具，这样你就可以直接看这些东西，并让人们知道寻找这些东西并识别它们。这是，呃，是的。这是一个难题。

And so like and it's in our libraries. And so you know like and we discovered this by accident like we saw this you know these chemical series keep popping up across all these different protein types it's like man like what is that is that captor and then you start looking at the protein expression and then like how it went it's like oh it it goes when when things aren't going well and so now we can use it as one filter for if we think if if we get a bunch of lumac heads that protein probably didn't fall fold right. We probably don't want to use that as training data. we probably want to rerun it with slightly different conditions and make it more friendly to the protein. Then we'll see what those hits look like. And so like unless you're paying that kind of attention like it's really I mean it's really hard. Yeah. And you have to build tools so that you can just look at these things and have people who know to look for these things and recognize them. It's uh Yeah. It's it's a hard problem.

是的。回到每个接触 Dell 的人出来都受到创伤。不知何故，你弄清楚了支持这个。

Yeah. Going back to everyone who touches Dells comes out traumatized. Somehow you've figured out supported this.

是的。你也受到创伤，但是以一种好的方式。是的。不，这真的很难。那就是，你知道，这就是为什么它，而且你知道，当你的动机只是找到一两个真正很棒的命中来开始一个药物项目时，那么你就不会追查像测定的所有这些不同方面。所以我理解。但如果你有动力让它成为机器学习的素材，那么就像是的，你必须注意所有这些事情。

Yeah. You're also traumatized but in a good way. Yeah. No, it's really hard. That's, you know, and that's why it's it's and and and you know, when you have the motivation to just finding like one or two really awesome hits to start off a drug program, like then you're not going to chase down like all of these different aspects of the assay. And so I get it. But like if you're motivated for it to be machine learning fodder, then like yeah, you have to pay attention to all this stuff.

我们最后总是问的一个问题是，如果你能通过法令消除一个瓶颈，那会是什么？

One of the questions that we always ask at the end is if you could by fiat remove a bottleneck, what would that be?

不。

No.

但当我想到 Bif FIA 时，它是目前正在发生的事情之一，就是看着测序成本像石头一样往下掉。我在想，好吧，我们真正想做什么？我们想要更多化学物质，我们想要更多测量。这些是我想要让我的生活稍微轻松一点的事情，就像那些眼前的问题。

But when I think about Bif FIA and it's one of these things that is actually currently happening, which is watching the sequencing costs drop like a rock. Like I'm thinking about, okay, what do we really want to do? We want more chemicals and we want more measurements. Those are the things that I want to make my life a little bit easier, like at the problems that are immediately on the ground.

所以你想要，抱歉我不明白。

So you want, sorry I don't understand.

更便宜的测序，更多分子。

Cheaper sequencing, more molecules.

好的。

Okay.

是的。

Yes.

在测序方面，你能给出一些背景，说明它下降了多少吗？

In sequencing, like can you give some context for how much it's dropped?

哦天哪，就像很多很多个数量级。我认为第一个基因组花费了十亿美元，现在我们可以用一千美元完成人类基因组，对吧？

Oh my god, like many many many orders of magnitude. It is, uh, I think the first genome cost a billion dollars and now we can do a human genome for a thousand, right?

我想可能是 100。

I think it might be 100.

真的很便宜。是的。有一段时间它下降得比摩尔定律还快。

Really cheap. Yeah. There were stretches where it was dropping faster than Mo's law.

是的。

Yeah.

那么等等，为什么你还想要更便宜的测序？

So wait, why then why do you want even cheaper sequencing?

哦，我只是，我只是对这个趋势感到高兴。

Oh, I just I'm just I'm just happy with the trend.

好的。如果你能通过命令移除一些东西，那会是什么？

Okay. What would you remove by fiat if you like that hasn't been removed?

是的，这里有个好东西。我想要的是，有不同的工具可以用来纯化小分子。好的。所以当你，呃，现在当你制造它们时，你把一些东西放在锅里，你加热或做其他条件，然后你得到输出，这会在某种效率水平上起作用。可能很好，可能很糟糕。所以无论哪种情况，特别是如果它很糟糕，还有这些其他工具可以用来清理它。所以人们使用像 HPLC 或其他一些技巧。它们中没有一个是超级可并行化的，真正非常好的。所以能够大规模清理东西会非常非常好，因为就像我们，你知道我们想做什么？我们在这种我们已经描述过的古怪测定中运行所有这些分子，这给人们带来了很多创伤。然后能在不同的测定中得到大量分子，这些分子会非常干净、纯化、易于处理，那就太好了，然后去做下一步，因为那样你就可以真正正交验证你正在做的所有测量。是的，那会非常非常好。当然，每个人都抱怨像 GPU 之类的东西。

Yeah, here's a good thing. What I would want is there are different gizmos that you can use to kind of purify small molecules. Okay. So when you, uh, right now when you make them you, you put some stuff in a pot and you, you heat it up or do some other conditions and you get the output and, and that will work at some level of efficiency. Might be great, might be crappy. And so in either case, especially if it's crappy, there are these other gizmos that you can use to kind of clean it up. So people use like HPLC or some of these other tricks. None of them are super parallelizable that are really really good. And so being able to like clean up stuff at a massive scale would be really really nice because like we, you know what do we want to do? We run all these molecules in this funky assay that we've already described has given people lots of trauma. It would be great to then get, you know, large numbers of molecules in a different assay that would be really clean and purified and easy to work with, uh, and then go do that next because then you could really orthogonally validate all those measurements that you're making. Yeah, that would be really really nice. And of course, everyone complains about like GPUs and stuff like

GPU，这是每个人说的，就像除了 GPU。

GPU that's what everyone says like besides GPU.

好的。所以，

All right. So,

然后另一个问题是，嗯，你希望人们从这一集里得到什么？

and then the other question is um what would you like people to take away from the episode?

我希望人们从这一集里得到的是，至少对我们来说，我们相信我们正在建造一些非常非常令人兴奋的东西，你知道，就像感觉像是物理基础的基底，用于各种有趣的推理和生物医学应用，所以如果他们对这个感兴趣，请联系我们，我们正在招聘机器学习人员，我们想和合作伙伴谈谈，所以你知道，我们正在建造这个我们认为会具有变革性的东西，所以来和我谈谈。

I would like people to take away from the episode that like you know at least for us we we believe that we're building something that's really really exciting and you know like what feels like physically grounded substrate for all sorts of interesting reasoning and biomedical applications and so if that's of interest to them like please reach out to us like we're you know we're hiring machine learning people we want to talk to partners and so you know and we're building this thing that we think will be transformative so come talk to me.

太棒了。

Amazing.

酷。

Cool.

那太好了。非常有趣。谢谢你来。

That was great. So much fun. Thank you for coming.

谢谢。是的。

Thank you. Yeah.

是的。

Yeah.