一粒沙中的世界：递归制药

我是凯·麦克拉里。

I'm Kay McClary.

这是马特·希利。

This is Matt Healy.

你正在收听《一粒沙中的世界》，我们将深入探讨一家最近在生物技术领域上市的公司的S-1文件（也称为招股说明书）。

And you're listening to the world in a grain of sand, where we dive into an s one or a prospectus, as they're called, of a company that recently went public within the biotechnology space.

今天，我们很幸运能讨论Recursion Pharmaceuticals，这是一家我们长期关注的公司。

Today, we're lucky enough to be talking about Recursion Pharmaceuticals, a company that we followed for some time.

这要感谢我们的一些同行在Lux Capital、Data Collective Ventures和Obvious Ventures的早期明智投资。

Thanks to early and sage investments of some of our peers over at Lux Capital Data Collective Ventures and Obvious Ventures.

这期节目很有趣，因为我们会充斥大量流行术语。

This is a fun one because we're going to be filled with a bunch of a bunch of buzzwords.

我觉得，麦克，Recursion在品牌塑造方面做得非常出色，你会听到诸如正交组学、恩维博米克斯、递归结、诱导实验室、超线性增长这样的词汇。

And I I feel like, Mac, to start off with, recursions done a great job of of branding, and you're gonna hear words like orthognomix, envybomix, recursion knots, induction labs, super linear growth.

我认为在S-1文件中，'激进'这个词至少被使用了50次。

I think that the word radical was probably used at least 50 times in the s one.

嘿。

Yo.

所有这些都说得对。

All of that is is is right.

我想我对这个有两点反应。

I I guess I had two reactions to that.

所以，首先，他们至少加了个术语表。

So so first, at least they put in a glossary.

对吧？

Right?

我想，认真说一句，我觉得他们术语表里的很多词，人们在各种场合都随意使用，但这些词的真实含义往往难以分辨。

I think, you know, to be serious just for a second, I think a lot of these terms that they put in the glossary, people throw around all the time in situations in which I think their true meaning isn't easily discernible.

我很欣赏他们加了术语表。

I really appreciate it that they put in a glossary.

我想，另一个对这些流行词的反应是，它们象征着，或者至少是代表了‘计算将解决药物设计与开发’这一整个理念。

I guess, and the other reaction to the buzzwords is they are the of the symbol or at least flag bearer for the whole idea of compute is going to solve drug design and drug development.

我们已经看到，凯恩，我简直数不清有多少份商业计划书都说，把一种药物推向市场要花费十亿美元，而我们要彻底改变这个行业。

And we've seen gosh, Kane, I can't even count the number of pitch decks that tell you it costs $1,000,000,000 to bring a drug to market, and we are going to transform the industry.

我们要降低这些成本。

We're gonna reduce those costs.

而所有这些主张的核心，都是以某种形式利用计算能力。

And at the at the center, at the crux of all of those claims is leveraging compute in some form.

是的。

Yeah.

其中一个

One of the

我认为他们最好的说法之一，是在我们深入探讨递归公司技术、使命和目标之前，即在最小化资金加权失败的同时最大化临床成功率。

ways one of the best ways I feel like that they they say it in the s one before we jump into the technology of the mission and goal of recursion apart from developing drugs that are gonna help many patients is to minimize the dollar weighted failure while maximizing clinical success.

这是一个宏伟的目标，但真正的希望在于，通过多种技术手段解析生物学，他们能够减少这种长期以来困扰生物制药行业的资金加权失败率——过去十到二十年间，基因组学革命带来了一些成功，但那些大多是低垂的果实或特定的基因靶点。

And now that's a that's a lofty goal, but the hope really is that by decoding biology through a number of techniques that will go into, they're able to minimize this dollar weighted failure function that has really plagued the biopharmaceutical industry over the last, call it, ten to twenty years with some some successes embedded in there from the genomics revolution that were essentially low hanging fruit or specific genetic targets.

但正如我们所知，生物学和疾病远比单一的点突变或单一靶点要复杂得多。

But as we all know, biology and disease is much more complex than, say, a single point mutation or single target.

因此，Recursion 公司的设立就是为了利用某种形式的计算能力来破解这些复杂性。

And so recursion was really set up to help us decode those complexities by using certain forms of compute.

这或许是一个很好的起点，来探讨 Recursion 的核心优势是什么。

That might be a good a good place to start in on on sort of what recursions sort of core advantage is.

对。

Right.

价值主张的关键，是我试图思考的方式。

The point of value proposition is the way I was trying to think.

比如，为什么 Recursion 比其他那些众多的机器学习、人工智能与药物发现结合的公司更具优势？毕竟这样的公司有很多，而且都有充分的理由。

Like, why is recursion better situated than any of the other, you know, kind of ML, AI meets drug discovery companies, of which there are many and and for good reason.

对吧？

Right?

是的。

Yeah.

我认为，他们最核心的优势在于我们 KDT 非常重视的一点，那就是成像和图像分析的力量，嗯。

And I think, you know, they're the the core advantage that they have remote centers around something that we at at KDT hold very near and dear, which is imaging and the power of image analysis, both Mhmm.

固定的图像，也就是单张图片，以及实时成像，你可以把它想象成一段视频，我们通过分析这些图像能做的事情。

Fixed images, so single pictures, as well as live imaging, which is more you can think of like a video reel and what we can do by analyzing images.

所以，让我们退一步来说，当我们想到图像时，尤其是生物图像。

And and so to take a step back, you know, when we think about images, they're really, or biological images specifically.

它们实际上是细胞内原子层面和蛋白质层面，以及基因组层面所发生事件的元数据集。

They're really a meta dataset of what's happening at the atomic and, say, protein level of cells and genomic level of cells.

因此，你通过图像获得了细胞是健康还是患病、是正在生长还是停滞、是否与其他细胞交流等形态学或基于图像的读数。

And so you're getting a a sort of morphological or image based readout of whether a cell is healthy or diseased, whether it's growing or it's fixated, whether, you know, it's talking to other cells or not.

图像可以帮助我们整合和汇聚所有这些分散的数据集，这些数据集可能原本被单独锁在遗传学、DNA或RNA测序、蛋白质组学，或者表观基因组研究中。

And images can help us start to collate and coalesce all of these disparate data sets that may have been locked up singly in genetics, in, you know, DNA or RNA sequencing or proteomics or, you know, whether we're looking at the epigenome.

实际上，这张图像是一种代理。

Really, that image is a proxy

代表了所有这些信息。

for all of that.

不。

No.

我觉得这是对的。

I I I think that's right.

我的一个体会是，当然我们会深入探讨，表型筛选是正确的方法。

One of my takeaways, and, of course, we'll dive in here, is is that the phenotypic screening is is the right approach.

如果这是衡量生物学和化学最强大的视角，那递归公司确实非常有优势。

And if that's the most powerful lens through which to kind of gauge and assess biology and chemistry, man, recursion's very well positioned.

是的。

Yeah.

是的。

Yeah.

所以，简单来说，从S1中我们可以提炼出递归的核心理念，这个名字也正是由此而来：我们可以在细胞中设置特定的操控或扰动，使其‘患病’。

And so, you know, the the basic idea of recursion as as we can sort of bring out of the the s one is the fact that and it it really drives that's where the name derives from, is that we can set up specific manipulations or perturbations in cells to make them, quote, diseased.

例如，对于单基因疾病，如果你的DNA中有一个碱基发生错误编码，我们就可以将这种突变引入细胞。

So if it's a monogenetic disease where you say have a single base of your DNA that is miscoded, we can introduce that into cells.

而且，我们特别可以将这种突变引入多种不同的细胞类型中。

And specifically, we can introduce that across multiple different cell types.

我们的身体由上皮细胞组成，比如皮肤细胞、胃肠道细胞，等等。

So our body, right, is made up of epithelial cells, your skin cells, your GI cells, things of that nature.

它还包含中胚层细胞，这些细胞处于中间状态，以及内皮细胞、成纤维细胞等各种不同的细胞类型。

It's made up of mesodermal cells, so things that that, lie sort of in in the middle ground as well as the endothelial cells and fibroblasts and all of these these different cell types.

因此，递归公司可以利用细胞培养方法来培养这些细胞，并开始引入这些扰动。

And so what recursion can do is if we have cell culture methods that allow us to grow these cells up, we can start to introduce these perturbations.

通过用多种试剂处理这些细胞，在递归公司的情况下，首先是他们的第一类药物或小分子，而且我们知道正常细胞应该是什么样子，或者拥有正常细胞的数据集，就可以观察到药物或化合物是否能让这些病变细胞恢复到正常的表型状态。

And by then treating these cells with a variety of agents, and in recursion's case, their first class of drugs or small molecules, and we know what normal cells are supposed to look like or have a data set of of normal like cells, you can start to see whether or not a drug or compound starts to recurse those diseased cells back to a normal phenotypic state.

所以，谁在乎它是如何起作用的，或者它具体在做什么呢？

So, you know, who cares about how it's working or what exactly it's doing?

我们是否能获得读数，显示那些曾经病变的细胞现在看起来更像、功能也更接近正常细胞？

Do we get the readout that shows us that something that was once diseased is looking much more and functioning much more like a normal cell.

这正是核心底层技术所在。

And so that's really the underlying core technology.

显然，这得益于递归公司内部强大的显微成像能力，以及你可以想到的其他使能技术，比如数据基础设施技术，无论是基于云的、图像分析技术，还是类似的技术，使他们能够以前所未有的规模实现这一目标。

And this obviously is enabled by immense microscopic power that recursion has brought in house, as well as you can think of other enabling technologies like data infrastructure technologies, whether they be, you know, cloud based or image image analytics or things of that nature, enabling them to do this at a scale that that really has never been tried before.

不。

No.

无论如何，我觉得这样说得很好，这也让我想到，是的。

That's I anyway, I think that's well stated, and and I think that brings me to yeah.

真相总是在最乐观和最悲观的观点之间。

The the truth is always somewhere between the most bullish view and the most bearish view.

而这里的乐观观点确实非常乐观。

And and so the bullish view here is is king pretty bullish.

对吧？

Right?

表型筛选结合最新的计算机视觉技术，是开发疗法的最佳平台。

It's that phenotypic screening powered by the latest advancements in computer vision is the best plane on which to develop therapeutics.

因此，Recursion 将成为最大、最成功的治疗公司，而且会从小分子开始。

And and as a result, you know, Recursion is going to be the largest and most successful successful therapeutic company, and they're going to start with small molecules.

他们明确表示，我们最终会进军生物制剂。

They expressly say, we're going to get to biologics.

我们会转向基因和细胞疗法。

We'll get to gene and cell therapies.

但让我们先摘取那些容易实现的成果，通过学习和训练我们的系统，持续收集数据，以便以他人无法做到的方式攻克复杂疾病，稍后我会详细说明。

But let us pick up the low hanging fruit, and we're going to learn and train our systems and continue to to gather data, and we'll get more into that here in a minute, to get to complex disease in a way that other folks can't.

这确实是一个非常乐观的前景。

And that's, you know, that's a pretty bullish case.

悲观的情况，至少在我看来，是这家公司许可了几个可能合理也可能不合理的小项目，并且在所谓很可能至关重要的基础设施层面上投入了大量资金，但我们仍不清楚。

The bear case, at least the way I think of it, is is this is a bear r and d arm that's in licensed a couple of programs that may or may not make a lot of sense and that has invested very heavily in in what is you know, what what what is likely to be important kind of infrastructure layers, but we still don't know.

我们仍然不清楚。

We still don't know.

而且，这同样是治疗领域和生物技术投资的一部分。

And, again, that's part of therapeutics, and that's part of biotech investing.

但你知道，我和你一样对表型筛选感到兴奋，主要是因为我们坚信计算机视觉的强大能力，这一点在我们的投资组合中随处可见。

But, you know, I as excited as I think you and I get around phenotypic screening largely because I think we are strong believers in in the power of computer vision, and you can see that across our portfolio.

你知道，现在仍处于早期阶段。

You know, the it's still early.

嗯。

Yeah.

嗯。

Yeah.

仍然非常早期。

Still really early.

问题是，如果细胞培养技术、显微镜和数据基础设施以某种方式对所有人开放，那么Recursion现在是否仅仅拥有先发优势？

And it's the the question is if if cell culture techniques, if microscopy, if data infrastructure is available to all in some way, shape, or form, does recursion only have that first mover advantage right now?

这仅仅是因为他们的基础设施吗？

And it's just is it just because of their infrastructure?

但他们的基础设施确实非常出色。

But their infrastructure is pretty freaking sweet.

这很酷。

That's pretty cool.

先发优势，我认为这是一个双刃剑。

And is the first mover you know, I I think first mover advantage is a is a dual sided sword.

对吧？

Right?

或者至少这是我理解的方式。

Or at least the way I think about it.

因为首先，你说的是，看。

Because because one, you're saying, look.

我是第一个行动的，这会形成某种护城河。

I'm first to move here, and that's gonna create some type of of moat.

但如果这真的是护城河，天啊，你最好完美执行。

But if if you're if that's really the moat, man, like, better be executing perfectly.

否则，我们的护城河正在迅速消蚀。

Otherwise, our moat is eroding as we speak.

所以我认为他们说得对，我们确实是第一家能获取如此多表型筛选数据的公司，这确实有其价值。

And so I think they're right that there is something to be said about we're the first company that's gonna acquire this many phenotypic screens.

但我并不认为搜索空间像其他领域那样小且可行。

But I don't know that the search space is as small, workable as it is in other spaces.

对吧？

Right?

比如Dino，他们的搜索空间是有限的。

Like, so so Dino, for example, that search space is is limited.

能进入衣壳的氨基酸只有那么多。

There are only so many amino acids that can go into A capsid.

衣壳。

A capsid.

这就像一个游泳池和一片海洋的区别。

It's like a swimming pool versus an ocean.

是的。

Yeah.

我不确定。

I I I don't know.

一方面，我理解这是先发优势。

I I'm I'm I'm I'm sympathetic on one hand to the this is the first mover advantage.

另一种看待方式是，好吧。

The other way of looking at it is like, okay.

所以，就像把印度洋划出来，其他人就会去抢占太平洋。

So, like, carve off the Indian Ocean, and someone else is gonna grab the Pacific.

我们可以转向南极，把其他所有海洋都拿下。

We can move down to the Antarctic and we can grab all of the other oceans.

所以，总之，这这

And so anyway, it's it's

这是一个有趣的观点。

a it's an interesting point.

我觉得，拜耳在这里做的本质上就是这样。

And that's essentially, I feel like what what Bayer did here.

所以，这是一个可以切入的点。

So that's a that's a place to jump in.

我知道我们还没讲完所有技术，但拜耳基本上说，嘿。

I know we haven't gone through all the technologies yet, but Bayer basically said, hey.

让我们从纤维化疾病这一领域中划分出一部分。

Let's carve off a portion of this ocean around fibrotic diseases.

这是Recursion签署的第二项业务合作。

And this was the second business development partnership that Recursion signed.

这是最近的一笔，也是最具意义的一笔。

It's the most recent one and and really the most meaningful one.

之前的合作伙伴是赛诺菲，早在2016年就达成了合作，并延长至2022年，但似乎这项合作并未取得太多成果。

The prior one was with Sanofi back in 2016, been extended to 2022, but it doesn't look like much fruit has born out of that partnership.

与拜耳的合作在几个方面出现了转折。

The Bayer partnership was inflected for a couple reasons.

首先，从交易条款来看，这看起来像传统的BioBuck合作模式，包含一笔前期付款和若干有限的临床项目，而所有这些项目都聚焦于纤维化或纤维化疾病，我认为这特别适合Recursion的平台。

One, I think from a deal term perspective, it looks like a traditional BioBuck partnership with some amount of upfront, some number of limited clinical programs, all of which though are centered around fibrosis or fibrotic diseases, which I think is particularly well suited for Recursion's platform.

成纤维细胞和成纤维细胞环境很容易培养。

Fibroblasts and fibroblastic environments are easy to culture.

由于这些细胞独特的形态特征，它们也容易分析。

They're easy to analyze because of the the particular sort of morphology of the cells.

因此，拜耳方面非常精明地率先获得了可能成为Recursion平台最容易实现的成果之一的权利。

And so it's a pretty shrewd part on Bayer's or Bayer's side of things to really start to get the rights on what could be one of the lowest hanging fruits on the recursion platform.

这笔交易的另一个关键点是，当我们进入这个阶段时，大约是2019年底的Recursion。

And the other inflected point of the deal was coming into this, so say late twenty nineteen, recursion.

当我们考虑药物开发时，没错，你有靶点发现，但你还需要新的分子实体。

So we we can't when you think about drug development, yes, you have target discovery, but you need new molecular entities.

对吧？

Right?

你需要新的化合物和化学物质。

You need new compounds and chemical matter.

Recursion在达成这笔交易时拥有大约几十万个化合物，当你考虑到化学空间几乎是无限的时，这个化合物库规模相对较小。

Recursion was coming into the deal with a couple 100,000 compounds, which is a fairly small library of compounds when you think about essentially infinite chemical search space.

拜耳并没有捐赠，而是作为协议的一部分，向他们提供了额外的50万个化合物。

And Bayer donated like, not donated, but as part of the agreement gave them 500,000 additional compounds.

于是，Recursion的化合物库规模一下子增长了三倍，这不仅对纤维化疾病领域至关重要，而且如果他们能将这些化合物用于其他筛选，并将通过这些项目从纤维化筛选中获得的数据整合到他们所谓的‘推断搜索’中——即无需实际将化合物作用于分子的计算机模拟搜索——那就更加重要了。

So all of a sudden, Recursion's compound library grew by threefold, and that's likely really important not only for the fibrotic disease set, but if they're able to use these compounds in other screens as well as embedding any data that they get from the fibrotic screens using these programs into what they call their inferential search, which is their sort of in silico search without actually putting compounds on the molecules.

说得好。

That's well said.

所以这里的看空观点是，这只不过是拜耳的一个研发合作。

That's why the bear case here is this is just a Bayer r and

d臂。

d arm.

是的。

Yeah.

70万个化合物中有50万个来自拜耳。

500 of the 700,000 compounds came from Bayer.

是的。

Yeah.

谁知道呢

And who knows

那个化合物库被筛选得有多彻底。

how picked over that library is to.

对。

Right.

所以，你知道，拜耳可能说过，嘿。

So, you know, Bayer might have said, hey.

我给了你们500千，但但但它们

I'm giving you 500 k, but But but they're

不是我最喜欢的。

not my favorite.

嗯。

Yeah.

它们并不是你知道的那些，或者也许是我们曾经 unsuccessfully 筛选过的那些。

They're not the ones you know, or maybe they're the ones we've mined unsuccessfully.

再次强调，正如你所说，这500个化合物的质量非常不明确。

Again, to your point, it's very unclear the quality of those 500 compounds.

有趣的是，他们内部所谓的化合物中有如此高比例来自拜耳。

It it's interesting that such a large percentage of their internal, quote, unquote, compounds come from Bayer.

是的。

Yeah.

没错。

Exactly.

好吧，让我们退一步，如果我们把化合物看作技术栈的一个层面。

Well, let's take a step back now if we think about compounds as sort of one lever layer of the technology stack.

所以，为了让你了解递归工作流程是什么样子，本质上是使用这些大型培养板。

And so to sort of walk you through what a recursion workflow looks like, essentially, use these really large plates.

它们是1536孔的培养板。

So they're 1,536 well plates.

你可以以某种重复方式接种你感兴趣的细胞。

You can essentially plate your cells of interest in some duplicate manner.

比如说，我们做三重复，或者每种特定细胞类型做10个正常样本，10个有扰动的样本，10个加药物一的样本，10个加药物二的样本，然后你就可以基于这些数据创建出非常庞大的数据集。

So call it we do them in triplicates, or we do 10 of each specific cell type in normal, 10 of them with perturbation, 10 of them with drug one, ten of them with drug two, and you can start to create these really large datasets off of that.

说到这个，他们现在的数据集每周大约增长80太字节，真是令人惊叹。

Speaking of which, their data set today grows at approximately 80 terabytes per week, so it's pretty incredible.

如果你开始算一下，本质上，Recursion 每周要处理大约 700 块这样的板子。

If you start to do some math, essentially, recursion is doing about 700 of these plates a week.

每块板子大约 500 美元，所以你可以合理地说，他们在板子筛选上每周烧掉约 30 万美元，因为他们每年大约运作 50 周，他们这么说的。

Each plate's about $500, so you can safely say they're burning about 300 k a week in plate screening, because they they function about 50 weeks a year, they said.

而在这些板子中，他们能够对细胞进行多种不同的离散组分或扰动。

And then within these plates, they they're able to do a number of different discrete components or perturbations to the cells.

首先，我应该说，这些都不是什么真正新颖的东西。

So one, there is and I should say none of these are really novel.

真正新颖的是它们的组合，以及其中的表型分析。

The thing is the combination of them and plus the phenotypic analysis therein is likely very novel.

第一个是遗传模块，包含 CRISPR 和向导 RNA 文库。

And so the first one is a genetics module, which consists of CRISPR and guide RNA libraries.

你可以想象，如果我想对细胞进行单碱基扰动，或者整个基因的敲入和敲除，我就需要一套 CRISPR 及其组件的文库。

You could imagine this as if I wanna perturb cells either single bases or sort of whole gene knock in and knockouts, I need a library of CRISPR and their components.

其次是一个可溶性因子模块，本质上是研究免疫调节。

Secondly is a soluble factor module, so essentially looking at immune modulation.

当你思考疾病状态时，疾病并不是孤立发生的。

So when you think about disease states, diseases are not happening in isolation.

它们发生在身体的基质中，这不仅包括该组织本身的细胞，还包括进出该组织的各类细胞，而其中有很多免疫细胞。

They're happening within the stroma of of the body, which includes not only whatever that resident tissue is, but whatever's traffic king in and out of that tissue, and that's a lot of immune cells.

因此，你需要能够调节这些细胞模型周围的免疫系统。

And so you need to be able to to modulate the immune system around these cellular models.

下一个模块是传染病模块，他们实际上会使用一些毒素，然后将它们加入到培养板的孔中。

The next one is they have a infectious disease module where, you know, they essentially take some toxins that are made and then can throw them into the wells of the plates.

下一个模块是纤维化模块，但他们并没有详细说明。

The next one's a fibrosis module, which they actually don't go into.

他们称之为正在开发中。

They call it under development.

这基本上就是拜耳在资助的项目。

That's essentially what Bayer is underwriting.

还有一个复杂的多细胞疾病模块，这其实只是说他们在单个孔中同时培养两种或更多种细胞类型。

There's a complex multicellular disease module, which is really just a fancy way of saying they do cold culture of two cell types or more in a single well.

这并不那么新颖。

It's not that that novel.

然后还有一部分是基于患者的，这部分发展得非常快，他们从患者身上获取疾病细胞，将其转化为诱导多能干细胞，然后能够将这些细胞分化成他们想要的任何不同细胞类型。

And then there's a patient derived side of the house, which is growing quite quickly, where essentially they take disease cells from patients, they then turn them into induced pluripotent stem cells, and then are able to differentiate those across whatever different cell types they want.

比如说，我们想获取患病的胃肠道细胞，将它们转化为iPSC，然后分化成成纤维细胞、神经元、内皮细胞等。

So let's say we wanted to take diseased, I don't know, GI tract, turn them into iPSCs, and then make them fibroblasts, neurons, make them endothelial cells, and you can differentiate them.

然后你可以想象，如果我把所有这些细胞都接种到培养板上并测试药物，不仅能了解药物是否能治疗疾病，还能知道治疗是否影响了体内其他细胞类型。

And then you can imagine if I then plate all of those things and test the drug, I can get not only indications on whether or not I'm treating the disease, but whether or not the treatment is affecting other cell types within the body.

这是一种相当有趣且新颖的方法，甚至可以用于临床试验的安全性评估。

And that's that's a pretty interesting and novel way to approach, say, even like a safety side of a trial.

对。

Right.

我印象深刻。

And I was impressed.

涵盖了65种疾病，近400个独立细胞系。

Nearly 400 individual lines across 65 diseases.

是的。

Yeah.

我的意思是，你看。

That I mean, look.

我知道他们已经拿出了 whatever it is。

I I know that they've taken whatever it is.

我们稍后会谈到。

We'll get to it.

四亿五千万美元的风险投资。

450,000,000 in venture capital.

这很多了。

That's a lot.

是的。

Yeah.

是的。

Yeah.

这太多了。

That's a lot.

看到在S1公司中，如此强调数据聚合和数据节奏，这在生物技术S1公司中并不常见，这非常有趣。

It's fascinating to see how much emphasis throughout the s one is put on the data aggregation and velocity of data cadence, which is not something you see all the time within biotechnology s ones.

因此，总的来说，我们可以开始思考如何利用这些大型平板来对细胞进行表型分析，希望找到一种有效的药物。

And so, you know, in general, we can start to think that we can use these large plates to start to phenotypically analyze cells, hopefully figure out a drug that works.

有趣的是，如果我们确实发现某种有效的药物，我们仍然不知道它为何有效。

Now interestingly, if we do see something that works, we still don't know how it works.

他们所指出的一个风险因素，特别针对Recursion的S1公司，是如果作用机制未知，他们可能难以获得IND批准。

And one of the risk factors that they outlined, which was specific to recursion in the s one, was that they may have trouble getting an IND if a mechanism of action is unknown.

而解析这种作用机制看似简单，实际上却相当复杂。

And decoding that mechanism of action may seem simple, but it's actually fairly complex.

因此，在这些检测的后端，他们不仅拍摄了细胞变化的图像，还配备了大规模的测序平台，以分析蛋白质组和基因组，从而了解当药物作用于这些疾病模型时发生了什么。

And so on the back end of these assays, they have not only the pictures that they're taking of what's going on, but they have a large sequencing suite as well to start to look at the proteome as well as the genome to see what is happening when these drugs are introduced to these disease models.

有趣的是，测序技术直到2020年才被整合进来。

Interestingly, the sequencing was only embedded in 2020.

嗯。

Mhmm.

所以有人会说，他们一直以来都是在外包，但有趣的是，他们在四年间筹集了4.5亿美元的风险资金，却直到去年才购买测序仪。

So one could argue that they've been outsourcing this entire time, but it seems interesting that they've raised $450,000,000 of of venture funds over four years and and didn't buy a sequencer until last year.

这让我几乎觉得我们之前漏读了什么。

So It almost it almost made me think that we weren't reading.

好像那里缺了点什么，是的。

Like, there was something missing there Yeah.

考虑到我，我认为测序仪——如果我们谈论我们的公司，每家公司都不同——这通常是有时甚至在A轮之前就购买的设备之一。

Given I I think Sequencer you know, if we're talking about our companies and every company is different, that's one of the pieces of hardware that is sometimes bought even before the a.

不是所有公司都一样，但这确实是非常早期的设备。

Not all we companies are different, but, like, that's a very early piece of equipment.

是的。

Yeah.

我猜Recursion会辩称，因为表型分析是核心，他们想先建立这方面的基础设施，测序则放在后面。

And my guess is that Recursion would argue that because the phenotypic analysis was core, they wanted to build that infrastructure out first, and then sequencing came later.

我认为，如果你开始审视他们所开展的项目，就会明显发现，我们仍处于Recursion平台可能发现成果的早期阶段。

And I think if you start to look at the programs that they have, it also becomes apparent that we're still in the early innings of what recursion may or may not be able to discover in their platform.

嗯。

Mhmm.

那什么是

And what

我的意思是，他们声称拥有37个内部项目，其中四个已进入临床阶段。

I mean by that is they claim 37 internal programs, four with four of which are clinical stage.

在这些项目中，只有一个被称作‘Recursion项目’。

Out of that, only one is a, quote, recursion

没错。

That's right.

项目。

Program.

没错。

That's right.

有些人可能会认为，这种药物其实早已在克里斯·吉布森的研究生研究工作中被发现，并非真正借助递归平台的力量发现的。

And some may argue that, you know, this drug was already kind of known from Chris Gibson's graduate study work and wasn't really discovered using the power of the recursion platform.

不。

No.

没错。

That's right.

因此，他们许可了多项这类项目。

So they've in licensed a number of these programs.

问题是，这是否会削弱平台的威力？

The question is, does that undercut the power of the platform?

我认为答案是我们还不知道，但这确实很有趣。

And I think the answer is we don't know, but it it's it's interesting.

我的意思是，如果平台一方面无法生成这些化合物，那就说明平台并没有我们所说的那么强大。

I mean, if the platform On one hand, would say if the platform isn't able to generate these compounds, then the platform isn't as powerful as we're saying.

当然，另一方面，他们还处于早期阶段，而且非常善于抓住机会。

The other side of the coin, of course, is they're early, and they were very opportunistic.

我不知道你有没有看过这些数字。

I don't know if you saw the numbers.

他们花了大约200万零1.5美元，还有125万美元。

They spent, like, 2,000,001.5, and 1.25.

这些是授权这些化合物所需的百万美元级别金额。

These are million dollar amounts to in license these compounds.

这可能很巧妙。

That could be crafty.

我的意思是，我们只能拭目以待，看看事情会如何发展，而且他们也拥有这些授权方的下游权利。

I mean, I like, we'll just have to see how this plays out, and they're a downstream right to these licensors as well.

但我对他们在授权这些化合物时如此精明的经济策略印象深刻。

But it's it I was impressed that they were so economically savvy on in licensing these compounds.

如果他们最终能进入临床阶段，我的意思是，他们确实做得很好，没花什么钱就拥有了多个项目。

If they end up getting to clinic, I mean, they did a really nice job of investing no money, and now they have programs.

即使只是上市了。

And even if it just got an IPO.

这就是我的意思。

That's what I mean.

没错。

Exactly.

他们的投入已经回本了。

Like, they this already paid for itself.

IPO估值是5亿美元。

The IPO was $500,000,000.

是的。

Yeah.

那我来简单介绍一下这些项目。

And so I'll do a quick rundown of the programs for us.

好的。

Great.

因为我觉得这些项目在很多方面都很有意思。

Because I think they're kind of they're interesting for a number of reasons.

不仅涉及它们的来源和目标，还可能包括临床试验中可能出现的一些障碍。

Not only sort of who they came from and what they're going after, but also maybe some some roadblocks coming up from the clinical trial perspectives.

第一个是REC4881。

The first one is REC four eight eight one.

REC4881是一种用于治疗家族性腺瘤性息肉病的药物。

REC four eight eight one is a drug for a disease called familial adenomatous polyposis.

本质上，这是一种由APC基因突变引起的单基因缺陷，患者胃肠道内会生长数十个甚至数千个息肉。

Essentially, this is a single gene defect in a gene called APC, and these patients grow tens, if not thousands of polyps in their GI tract.

其中许多息肉具有不典型增生，会发展为癌前病变，最终演变为癌症。

Many of these are dysplastic and turn into precancerous and eventually cancerous lesions.

这种特定化合物是从武田公司获得授权的。

This specific compound was licensed from Takeda.

正如马克斯所说，前期付款为150万美元。

As Max said, it was a 1,500,000.0 upfront.

还有3950万美元的里程碑付款，如果最终成功上市，还需向武田支付个位数的版税。

There's 39,500,000.0 in milestones, and then single digit royalties owed to Takeda if they end up taking this all the way.

现在，在他们的S1中，4881详细介绍了让他们有信心将其推进临床的数据。

Now 4881 in their s one, they detail the data that got them comfortable with taking it into the clinic.

当你查看这些数据时，他们给小鼠的剂量在每公斤一到十毫克之间。

Now when you look at the data, they are treating these mice at somewhere between one and ten milligrams per kilogram.

所以在每公斤一毫克的剂量下，效果与塞来昔布差不多，而塞来昔布本质上是一种这些患者正在服用的抗炎药。

So it works for at one milligram per kilogram, not that much better than celecoxib, which is essentially an anti inflammatory that these patients are taking.

在每公斤十毫克的剂量下，效果相当显著。

At ten milligrams per kilogram, it's fairly profound.

假设一个成年人体重为七十公斤，这是平均体重。

Now let's say that a human is seventy kilograms, average human.

如果你想达到每公斤十毫克的剂量，就需要给他们服用七百毫克。

If you wanted to hit the ten milligram per kg, you're gonna be giving them seven hundred milligrams.

如果你想达到每公斤一毫克的剂量，就需要给他们服用七十毫克。

If you wanna hit the one, you're gonna give them seventy milligrams.

Recursion公司以十六毫克的最大耐受剂量开展了第一阶段临床试验。

Recursion ran the phase one trial at a maximum tolerated dose of sixteen milligrams.

所以远低于每公斤一毫克的剂量。

So well below even a one milligram per kilogram.

对。

Right.

因此，问题是，当他们继续推进到二期和三期临床试验的关键阶段时，这里的剂量是否过低了？

And so the question is, as they continue to move into the actual pivotal part of the trial in phase two and phase three, are they undershooting here?

这可能是其中一个经典问题：我们能治愈小鼠，却无法治愈人类，因为副作用谱的差异。

And this may be one of those classic or canonical problems where we can cure a mouse, but can't cure a human because of the side effect profiles.

这是一个非常值得追踪的项目。

It's a really interesting program to track here.

我们希望并支持这个项目能为这些患者带来疗效，但确实存在一些潜在的障碍。

We hope and we're rooting for the the program to work for these patients, but there are some interesting potential roadblocks.

第二个是REC 3,599。

The second one's rec 3,599.

这专门针对GM2神经节苷脂贮积症，即泰-萨克斯病，本质上是一种代谢性疾病。

This is specifically in a GM2 gangliosidosis, which called Tay Sachs disease, which essentially is a is a metabolic disease.

他们以125万美元的前期费用、7500万美元的里程碑付款以及个位数的版税从Chromoderm获得授权。

They license this from Chromoderm for again, one point two five million upfront, seventy five million of milestones, single digit royalties.

这真是个精明而廉价的举措，能获得一个尚无体内数据、但临床前研究看起来相当有前景的项目。

So shrewd cheap move to get something that still has no in vivo data, but is looking quite robust from a preclinical standpoint.

作为提醒，这两项研究预计都将在明年三月左右启动二期临床试验。

And both of these, just as a reminder, expected to launch the phase two within the next, you know, March.

是的。

Yeah.

仍然处于早期阶段。

Still nascent.

同样，我们只是种子投资者。

Again, not not and we are seed investors.

这并不是一次成功，但它提醒我们，公司在何种阶段仍有可能实现成功的推出。

That is not a hit, but it's it's a reminder of where companies can be and still have you know, what what's been a successful launch.

对吧？

Right?

一次成功的IPO。

A successful IPO.

嗯。

Mhmm.

他们经历了一些起伏，但市场也经历了起落。

They've had some ups and downs, but the market's had some ups and downs.

我认为人们会称这到目前为止是一次成功的IPO。

I think I think folks would call this a successful IPO so far.

还差得远呢。

Still really far.

对吧？

Right?

比如，这四个临床项目，难道不都预计在接下来的400天内达到二期或三期吗？

Like, still, all of these four clinical programs, aren't they all supposed to be phase at best, it's phase two and three within the next four hundred days?

是的。

Yeah.

我认为有一个项目将直接进入二期和三期，其他三个项目则将在明年某个时候启动二期。

I think there's one that's going to do a phase two and three, then the other three are just start phase two sometime next year.

这还非常早期。

It's really early.

这应该会为那些更早阶段的公司带来一些推动或鼓舞，让他们思考：我们的公共资金是否可及？是的。

That should give, again, kind of a kick in the step or some enthusiasm to those companies that are about that stage even earlier as they think about, you know, are we are public funds accessible Yeah.

对我们来说。

To us.

当然。

Absolutely.

下一个项目是Rec2282。

So the next one's Rec two 282.

这是一种从俄亥俄州立大学授权的神经纤维瘤病2型药物。

This is a neurofibromatosis type two drug licensed from Ohio State.

有趣的是，俄亥俄州立大学原本在开发这种化合物作为癌症治疗药物，但这种药物能穿透中枢神经系统，作用于组蛋白去乙酰化酶。

Interestingly, this was Ohio State was pursuing and is still prosecuting this compound as a cancer treatment, but this is a CNS penetrant drug that works on histone deacetylases.

同样，俄亥俄州立大学的许可看起来和其他所有许可一样：200万美元的前期费用，2000万美元的里程碑付款，以及个位数的版税。

Again, the Ohio State license looks like all the other 2,000,000 up front, 20,000,000 milestone, single digit royalties.

同样，动物实验结果与人类耐受性之间存在着有趣的矛盾。

Again, an interesting dichotomy behind what has been done in animals and what we know is tolerated in humans.

因此，S1中展示的所有动物研究都是在每公斤25毫克的剂量下进行的。

So, the animal studies that were all presented in the s one were at twenty five milligrams per kilogram.

这相当于人体大约1.4克左右，但人类实际使用的剂量仅为30至60毫克。

So that would be, you know, one point four grams give or take for a human, but humans are only getting thirty to sixty milligrams of this drug.

那么，这个剂量对神经纤维瘤病2型（NF2）来说足够吗？

So again, is that enough for NF2?

谁知道呢？

Who knows?

所以，这些就是我们引进的三个项目，目的主要是为了跳过排队等待。

So those are our three programs that we licensed in to basically jump the queue.

最后一个项目是脑海绵状血管畸形。

And the last one is a cerebral cap cavernous malformation.

所以是Rec994。

So Rec nine nine four.

Recursion公司在这里进行了第一阶段试验。

Recursion did the phase one here.

他们在这里拥有临床试验的专业能力。

They have clinical trial expertise here.

这可能是我们刚才提到的三个适应症中最大的一个。

It's likely the largest indication of the three that we just mentioned.

有趣的是，它被严重漏诊，这一点在他们的S1文件中被明确指出了。

And interestingly, it's severely underdiagnosed, and that's brought out pretty loudly in their s one.

然而，他们完全没有提及任何伴随诊断工具，或任何能提高诊断率以在药物获批后吸引市场的策略。

However, there is no talk about a co diagnostic or something to increase the diagnostic rate to essentially pull the market toward them if they have drug approval.

因此，每次开发药物时，这一点都值得深思。

So something to think about there every time you're developing the drug.

如果某种疾病被漏诊，你需要弄清楚原因，并最好在开发药物的同时，嵌入某种诊断方法或项目，以将患者或市场吸引过来。

If something's underdiagnosed, you need to figure out why and ideally start to embed some diagnostic paradigm or program alongside the drug that you're developing to bring those patients or market to you.

这是一种有趣的药物，它的作用机制很特别。

It's an interesting drug and how it works.

真正的问题是，这些患者在确诊时，大脑的动静脉系统已经出现了严重的畸形。

The question really is these these patients, by the time they're diagnosed, have these pretty bad malformations in the arterial venous system in their brain.

这种药物真的能重新构建这些结构吗？

Is this drug really gonna re architect those things?

它只是缓解症状吗？

Is it just gonna treat the symptoms?

这又引出了作用机制的问题：

And again, this portends to that mechanism of action question of

嗯。

Mhmm.

这种药物在这一特定疾病中的确切作用机制是什么，以及他们观察到的表型是否能转化为成功的临床项目。

How exactly is this going to work in this specific disease set, and whether or not what they were seeing phenotypically is gonna translate to to a successful clinical program.

目前还有多项先导化合物优化正在进行。

There's a number of other lead optimizations going on.

这些都很有趣，但都是通过他们所谓的暴力实验项目完成的，他们再次将化合物喷洒在样本上，观察哪些化合物能让细胞最接近恢复正常状态，但他们最引以为傲、也最常提及的是两个新的推断性项目。

Those are all interesting, but they were all done by their quote brute force program, where again, they're spraying their compounds on it and seeing which compounds recurs the cells closest back to normal, but they're most proud and I think most loud about the two new inferential programs.

对。

Right.

在你深入这些之前，肯，对于那些还没读过这份S1文件的人，递归公司明确指出了一个清晰的演进过程：历史上，我们一直使用暴力实验法，而暴力实验本质上就是通过高通量筛选平台直接测试大量试剂扰动组合，对吧？

And just before you get into that, Ken, just for the folks that haven't read this s one, there's a pretty clear progression that recursion cites, it says, look, historically we've been using brute force, and brute force is essentially direct experimentation, right, to evaluate large number of reagent perturbation combinations through their high throughput screening platform.

这说得通，对吧？

So that makes sense, right?

这相当于把传统药物发现方法放大了无数倍。

That's historical drug discovery kind of on steroids.

他们在S1文件中反复强调，这是一种历史做法，而他们现在已经转向了推断性方法。

They make it very clear throughout the S1 that that's a historical practice and that they've moved forward to inferential.

推断性方法本质上是一种利用机器学习预测未来结果的方式。

Inferential is essentially a way to discover ML enabled predictions for future results.

对吧？

Right?

这是在提供充足数据的情况下，计算所能做出的推断。

It's an inference that compute can make when when provided with with adequate data.

真正有趣的是，需要进行的实验数量与治疗和疾病扰动库的大小之和成正比，这让我觉得数量实在太多了。

And what's really interesting is that the number of experiments to be conducted is proportional to the sum of the sizes of the therapeutic and disease perturbation libraries, which to me just makes me think that's a lot.

这真的很强大。

Like, that is powerful.

当我退后一步思考时，我认为他们大力宣传这种推断能力是有充分理由的。

When I took a step back and I thought there's a good reason for them to, I think, trumpet this inferential capability.

对吧？

Right?

因为这再次回到起点，正是人们一直试图推向市场的东西。

Because that is again, going back to the start, that is what folks have been trying to bring to market.

它能够将人类从所有偏见以及无法客观、高效地分析数据的局限中剔除出去。

It is the ability to take humans out of all of our biases, all of our inabilities to actually think analyze data objectively and in in a high throughput manner.

发现那些

See relationships that

我们太笨了，看不到。

we're too dumb to see.

没错。

Exactly.

这就是推理程序。

That is the inferential program.

所以，总之，在你进去之前

So anyway, before you go in

如果你有这些化合物，我觉得这再完美不过了。

I think it's perfect if you have the compounds.

是的。

Yeah.

我仍然在思考关于递归的问题，当总的化学库达到几十万种化合物时，如果我们把拜耳的化合物也包括在内。

And I'm still wondering on the recursion side of things with a total chemical library in the 100 high hundreds of thousands, with we include the Bayer compounds.

他们声称拥有数十亿级别的计算机模拟库。

They claim an in silico library in the billions.