Can AI Help Solve Alzheimer’s?

这是iHeart出品的播客《 Guaranteed Human》。

This is an iHeart podcast, Guaranteed Human.

很多职场建议听起来头头是道，可真到用的时候就全崩了。

A lot of work advice sounds good in theory, but falls apart when you actually try to use it.

我就是要改变这种状况。

I'm on a mission to change that.

我是莫莉·格雷厄姆，创业者，也是TED旗下《Work Life》节目的新主持人。

I'm Molly Graham, a company builder and the new host of Work Life, a podcast from TED.

我职业生涯都在高速成长的公司里打滚，我最确定的一件事就是：工作从来就不是井井有条的。

I've spent my career inside fast growing companies, and one thing I know for sure is that work is messy.

这一季，我特别想和你聊聊那些创始人、操盘手和创意人——他们最光鲜的成功背后，那些没人会在领英上晒出来的真事儿。

In this new season, I'm excited to share my conversations with founders, operators, and creatives about the real story behind their shiniest successes, the lessons that no one ever posts on LinkedIn.

现在就去你常用的播客平台，收听《Work Life》吧。

Listen now on Work Life wherever you get your podcasts.

Pushkin出品。

Pushkin.

你知道摩尔定律吧。

So you know Moore's law.

对吧？

Right?

就是说电脑，特别是芯片，性能越来越好，价格越来越低，而且是指数级提升。

The idea that computers or specifically chips get better and cheaper at an exponential rate.

搞新药研发的人会提到一个叫‘逆摩尔定律’的东西。

People who work on researching and developing new drugs, they talk about Eroom's law.

‘Eroom’就是‘Moore’倒着拼，这说法半是玩笑半是认真，意思是几十年来，新药研发变得越来越慢、越来越贵。

Eroom is more spelled backwards, and this is sort of a half joke, half not a joke way of pointing out that developing new drugs has gotten slower and more expensive over the past several decades.

这跟芯片的发展方向完全相反。

It has gone in the opposite direction of developing new microchips.

现在，尽管在新药研发上投入了这么多钱、技术、人力和智慧，但进入临床试验的药物里，差不多有百分之九十都失败了。

Today, after all the money and technology and hard work and intelligence poured into drug development, something like ninety percent of drugs that go into clinical trials fail.

只有百分之十能成功。

Only ten percent succeed.

也就是说，如果我们能让成功率提升到百分之二十，哪怕失败率还是百分之八十，我们的进展速度也能翻倍。

That means that if we got to a place where even just twenty percent succeeded, still an eighty percent failure rate, we would double the rate of progress.

我是雅各布·戈德斯坦，欢迎收听《你的问题是什么？》

I'm Jacob Goldstein, and this is What's Your Problem?

今天做客的嘉宾是帕特里克·许。

My guest today is Patrick Hsu.

帕特里克21岁就从哈佛拿到了生物化学的博士学位。

Patrick got a PhD from Harvard in biochemistry when he was 21 years old.

四年前，还没满30岁，他就创办了Arc研究所。

And then four years ago, before he turned 30, he founded the Arc Institute.

这是个位于湾区的非营利研究机构，聚集了来自斯坦福、加州大学旧金山分校和加州大学伯克利分校的科学家，帕特里克本人也在伯克利任教。

It's a nonprofit research center in the Bay Area that hosts scientists from Stanford, UC San Francisco, and UC Berkeley, where Patrick is on the faculty.

帕特里克的核心问题是：怎么用人工智能让生物研究更高效，更快地帮科学家找到能带来新药和更好疗法的突破？

Patrick's problem is this: How can you use AI to make biological research more efficient, to guide scientists more quickly to discoveries that'll lead to new and better treatments?

稍后你们会听到，帕特里克特别关注阿尔茨海默病。

As you'll hear later in the conversation, Patrick is particularly focused on Alzheimer's disease.

一开始，他跟我讲了为什么基础生物学研究到现在还是这么慢。

To start, he told me about why basic biological research is still so slow.

你要是看看八九十年代、两千年初，或者2010年代初，再到今天的2020年代，生物实验室的样子，基本没变。

So if you look at a biology research lab in the eighties or the nineties, the early two thousands, the early twenty tens, or today in the twenty twenties, they look basically the same.

对吧？

Right?

那儿有长长的实验台。

You have these long benches.

还有两三排架子，微量移液器，一堆机器，看着跟家用厨房用具似的，对吧？

You have these two or three rows of shelves, these micro pipettes, and various machines that look like home kitchen equipment, right?

所以研究生、博士后或者实验员，就像米其林餐厅里的灶头厨师。

And so graduate students and postdocs or bench scientists generally are like the line chefs, right, inside of a Michelin starred kitchen.

你负责备菜、调基础酱汁。

You're prepping the vegetables, you're making the sort of initial sauces, right?

比如取组织、处理细胞，再用抗体染色什么的。

So you might be taking tissues down or cells, processing them, staining them with antibodies.

我觉得关键在于，这些实验特别慢，全靠手工操作，还特别依赖那些说不清道不明的软性经验和技巧，而且每个实验室的做法都不一样。

I think the point is that doing these experiments is extremely slow, very manual, and requires a huge amount of soft knowledge and know how, and is really variable lab to lab.

所以我们面临的是一个多层难题——得一步步去探索，对吧？

So we have this multi stack problem where we have multi step search, right?

实验动不动就要花几个月甚至几年，我们基本上全靠手动一点点去试遍所有可能性，对吧？

The experiments take months to years to run out, and we basically search across the entire space in an extremely manual way, right?

在如今这个AI能预测很多领域事物的时代，问题来了：我们能不能靠有预测能力的模型，来加速生物学研究？

And so in the modern era of AI, where we believe we can predict things across many different fields, the question is, can we speed up biology by being able to have models that actually have predictive value and power?

我们的目标是打造一种生物智能，让细胞生物学家能用模型来筛选出最值得做的12个实验，而不是像现在这样，凭假设一头扎进实验室瞎试。

And our goal is to be able to create biological intelligence that starts to crack that threshold of a cell biologist will use the model to rank the top 12 things that they'll do in the lab, rather than just going to lab and trying a bunch of things based on hypothesis driven science.

而是先靠模型预测，再进实验室验证。

But you actually do model first prediction and then a lab in the loop experiment.

当然，这些前提条件我们待会儿再细说。

And, I mean, we'll get to the sort of things that have to happen for that to happen.

但要是真能做成，那可就太棒了。

But But if we could get it to work, Yeah.

如果真能成，那除了实验室之外，还能带来什么更大的好处？

If you could get it to work, what's the payoff beyond the lab?

我觉得首先，做实验的设计和操作方式会彻底改变，对吧？

I think the first thing is the practice of how you design and do experiments will completely change, right?

就像现在研究生想了解一个新领域，直接拿ChatGPT问一问就行，对吧？

And so just like, you know, graduate students today, whenever they want to research a new area, they just use ChatGPT, right?

不管是Claude、Gemini，还是你用得顺手的其他模型，都成了标配。

Or Claude or Gemini or whatever your favorite model is, right?

这已经成了科研流程里再自然不过的一环。

That has just become a fundamental part of the workflow.

其实我觉得，现在大多数人早就不再像以前那样读论文了，对吧？

In fact, I don't think most people read papers the old fashioned way anymore, right?

以前是把论文从纸质杂志搬到浏览器里看，比如《自然》《科学》。

And it used to be, oh, I read them online in my browser instead of in a magazine like Nature and Science.

现在呢？直接把PDF丢进ChatGPT，看它给的总结就行。

And now it's just, I throw the PDF into ChatGPT and I read the summary.

对吧？

Right?

这跟几年前的实验方式完全不一样了。

That's just a completely different workflow than just even a few years ago.

我觉得以后我们不再亲手设计实验，而是让AI帮我们设计、排查问题，决定下一步该怎么做。

And I think instead of designing experiments by hand, we're gonna have AI models design the experiments for us and help us troubleshoot and decide what to do next.

我听你把现在的生物学研究方式形容成‘猜了再试’，对吧？

I've heard you describe the way biology works now as guess and check Mhmm.

我觉得这说法挺好，某种程度上确实就是科学方法嘛。

Which I liked, which I guess on one level is kind of the scientific method.

对吧？

Right?

其实就是假说加实验嘛。

It's kind of like hypothesis and experiment.

但这个‘猜’字，恰恰说明了其中的不确定性，那些我们走过的弯路和死胡同。

But the guess really leans into the, not quite blind, but the uncertainty, all the blind alleys that, I guess, you go down.

没错。

Exactly.

这就跟小时候玩的《战舰》游戏差不多，你得想办法击沉对方的战舰，但又不确定它到底在哪。

You you know, it's a lot like that that childhood game Battleship, where you're trying to sink someone's Battleship, you don't quite know where it is.

所以你得一个个区域去搜，就像《扫雷》那样，对吧？

And so you're searching these different quadrants, or like Minesweeper, right?

根本很难知道你到底有没有找对地方。

It's just very hard to know that you're searching in the right place.

所以第一步是，这些AI模型能不能帮我们找准方向？这样我们再做那些手动的湿实验时，成功率就能大大提高——不管是设计药物，还是预测细胞对某种扰动的反应。

And so the first thing is, can these AI models help us search in the right place so that when we're peppering things with these individual manual wet lab experiments, our hit rate can just significantly go up, whether that's for designing a drug or, you know, predicting how cell will respond to, you know, some, perturbation.

对。

Yes.

这算是最基础的第一步。

And then that's kind of the first order thing.

说白了，就是让基础研究变得更高效。

So it's basically making basic research more efficient.

速度。

Speed.

对。

Yeah.

不过呢。

But yeah.

那第二、第三层的影响是什么呢？

What are the second and third order things?

比如，这怎么转化成临床结果？

Like, how does that translate to clinical, outcomes?

我觉得它们是直接相关的，对吧？

I think they're directly linked, right?

是的。

Yeah.

在药物研发中，我们会做一件事，叫靶点识别，对吧？

In drug discovery, we do a thing called target identification, right?

你得能找到正确的药物靶点，还得知道怎么精准地干预它。

You need to be able to find the right drug target, and you need to be able to perturb it in the right way.

你要找出出问题的那个东西——比如导致疾病的毒性蛋白，然后想办法把它关掉，对吧？

You need to find the thing that's going wrong, the toxic protein that's causing a disease, and you try to turn it off, right?

过去我们的目标一直很单纯：找到对的靶点，再用对的方法去药理干预。但问题是，对大多数阿尔茨海默病、多种癌症、衰老、自身免疫病这些主要致死疾病，我们根本搞不清靶点到底是什么，连猜都猜不准，情况到现在还是一样。

So the goal has really been just finding, the right drug target and then drugging in the right way, but the problem is we don't seem to find the right drug target or even know what it is for most Alzheimer's complex disease, many different types of cancers, aging, autoimmunity, a lot of the major killers, it's still the same.

我们其实根本不知道该去靶向什么。

We know what to actually target.

我们觉得这些模型能帮我们解决这个问题。

We think these models could help us do that.

咱们聊聊Arc研究所吧，很多这项工作都是在那里做的，是你创办并负责的机构。

Let's talk about the Arc Institute, which is where a lot of this work is happening, a thing you've started and you run.

你给说说，Arc研究所到底是怎么回事？你当初为啥要创办它？

Tell me about the Arc Institute and, like, specifically why you started it.

Arc研究所今年刚满四年。

Arc is about four years old today.

我们团队现在大约有350人，是一家全栈式的AI与生物研究机构。

We're about 350 people, and we're a full stack AI and biology research organization.

也就是说，我们把实验科学和计算科学真正融合在一起，同在一个屋檐下工作，让AI模型的训练和运行能跟实验室的实验无缝循环：一边跑模型，一边在实验室里验证，不断迭代，从而发现新的作用机制、开发新药、找到新的治疗靶点。

And so by that, I mean, we try to natively combine experimental science and computational science under a single physical roof so that we can do native iterative lab in the loop between AI models, both training and running them, and then actually experimentally validating and verifying things in the lab in order to create new mechanistic insights, new drug compositions, and identify new therapeutic targets.

我们有两个主要目标。

We really have two major goals.

第一个是打造虚拟细胞模型，用基础模型模拟人类的生物学过程。

The first is to create these virtual cell models that can simulate human biology with foundation models.

第二个目标是治愈阿尔茨海默病。

The second is to cure Alzheimer's disease.

我们希望Arc能像爱迪生的实验室那样，对吧？

We're interested in the view of Arc as a sort of Edison shop, right?

就是能一口气发明一堆东西，然后真正做成产品，对吧？

Where you're inventing a whole bunch of different things that we can actually productize, right?

所以，Arc的野心已经不只是做突破性的学术研究了，而是想把科学真正变成能用、能看、能摸、能吃能用的东西，让普通人实实在在受益，而不仅仅是发几篇论文。

And so in a way, ambition from Arc has been increasing beyond how do we do breakthrough academic and basic science to how do we actually productize this science in a way that we can impact human health, not just with research papers, but with things that people can actually take, see, and feel, and use.

你说的爱迪生实验室，那本来是个商业机构，对吧？不是慈善组织。

When you talk about the Edison shop, I mean, obviously, the Edison shop was a was a business, right, not a philanthropy.

那你现在是怎么考虑商业化这条路的？

Like, are you how are you thinking about it commercially?

Arc 是个非营利机构。

So Arc is a nonprofit.

对吧？

Right?

是的。

And so Uh-huh.

Arc 研究所会一直保持这个非营利的、以发现为导向的母体，但我们也会有能力、也有意愿孵化出一些独立实体，去拿更聚焦的商业资本，把成果真正做成产品。

You know, I I Arc will the Arc Institute will always remain this nonprofit discovery oriented mothership, but we also will certainly have the capacity and and the interest to be able to spin out entities that can take on more focused commercial capital in order to productize things.

咱们还是回头聊聊你们做的机器学习和 AI 工作吧。

Let's go back to the machine learning slash AI work you're doing.

我觉得咱们该谈谈 Evo。

I think we should talk about Evo.

给我讲讲Evo吧。

Tell me about Evo.

Evo其实就是个ChatGPT，只不过它处理的是DNA。

EVO is essentially a chat GPT, but only with DNA.

对吧？

Right?

明白了。

Okay.

输入的是DNA，输出的也是DNA。

And so it's DNA in and DNA out.

对吧？

Right?

你可以用核苷酸跟它对话，它会返回一串对应的核苷酸序列。

So you can talk to it by typing in nucleotides, and you'll receive from the model some corresponding set of nucleotides.

就像你要是说‘生存还是毁灭’，它就会接下去说‘这是个问题’，懂吧？

So just like I could say, to be or not to, the model would tell me, know, you know, to be or not to be, that is the question, you know?

然后它会继续下去，把《哈姆雷特》那段独白的其余部分都补全，对吧？

And then it'll keep going, and the rest of Hamlet's soliloquy, right?

如果我给它一段线粒体基因组，或者比如说一个E.

If I gave it a fragment of, let's say, a mitochondrial genome or, you know, an E.

大肠杆菌的基因组，它就会试着自动补全剩下的部分——这可不是简单地记忆和复述训练数据，这才是关键，对吧？

Coli genome, it will then try to auto complete the rest of It's that not just memorizing and regurgitating information from the training data, which is what's really important, right?

它是在做一种语义上的延伸，根据它理解的含义，去生成它认为应该生成的东西。

It's doing a semantic diversification in a way of what, yeah, what it thinks is in the meaning of what it needs to make.

对。

Yeah.

最神奇的就是这一点，对吧？

That's the wild part, right?

如果只是复述训练数据，那根本算不上人工智能。

If it was just regurgitating training data, it wouldn't be AI.

那不就是个数据库嘛，对吧？

It would just be sort of a library, right?

没错。

Exactly.

这就是为什么你可以让ChatGPT帮你写邮件，比如你跟它说：‘这是我希望邮件里写的，你帮我写一封。’

And this is why you can ask ChatGPT to, you know, here's what I want to see in my email, write me the email.

就算你这么跟它说十遍，它每次写的邮件都会差不多，但又不完全一样。

And if you gave it that thing 10 different times, it would write you similar, but different emails.

对吧？

Right?

是的。

Yeah.

同样地，你也能设计出不同的CRISPR系统，或者不同的线粒体基因组——它们很相似，但又各有不同。

And just similarly, you can make different CRISPR systems that are similar, but not the same different mitochondrial genomes that are similar, but not the same.

你可能会说，哎，这挺有意思的。

And you might say, okay, well, that's fun.

现在你电脑里就有一堆序列了。

Now you have a bunch of sequences in a computer.

有趣的是，我们可以在实验室里用化学方法合成这些DNA链，然后测试它们的功能。

What's The interesting about point is that we can then chemically synthesize those DNA strands in the lab and test them and see what their function is.

对。

Uh-huh.

嗯，没错。

Uh-huh.

接下来，我们可以把实验室测出的数据反馈给模型，告诉它：‘帮我优化这个实验任务’。这样一来，我们就形成了一个闭环——能不断迭代，提升任何你关心的实验效果。

Now, we can then take those, the numbers that we've measured in the lab and feed it back into the model and then tell the model for things like make it better at this task that I've experimentally And then you can this is the closing the lab and the loop that we've built Arc around in order to hill climb on really any experimental assay that you care about.

我记得去年你一位同事做过一个项目，用EVO2设计了一种新的噬菌体。

I I know there was a project that, one of your colleagues did last year using EVO two to basically to to design a new version of an existing phage.

噬菌体就是感染细菌的病毒。

A phage is a is a virus that infects bacteria.

能说说这个φX174噬菌体的项目吗？它为什么重要？

So tell me about the project with the phi x phage and why it's meaningful.

我们有个数据库，只要有人发表论文，就会把基因序列上传到里面。

So we have these databases where whenever people publish papers, they deposit the genetic sequences into this database.

所以对于phi x以及相关的噬菌体，我们目前已经能大致把握住它们在自然界中的进化分布情况。

So of all phi x and phi x related phages, we have a sense of the evolutionary distribution that is out there that we've been able to detect so far as scientists.

结果发现，EVO能创造出全新的phi x版本，其进化差异程度跟我们见过的其他所有类型不相上下。

It turns out EVO is able to make new versions of PHY X that are as evolutionarily distinct as everything else that we've seen.

我们可不是在简单地复制，搞出什么凯美瑞或者雅阁的翻版。

So we're not just copying and making things that look like a Corolla or another Accord.

那太没意思了，我们能做出一种完全全新的车型。

That would be very boring, but we can make, you know, a fundamentally new type of car.

对吧？

Right?

没错。

Yeah.

它还是车，但设计思路完全是全新的。

It's still a car, but it's a completely new type of design.

比如说，像PT Cruiser那样独一无二的车型，但还要酷上一百倍。

Let's say something, like, as unique as a PT Cruiser, but much, much cooler.

而且

And

那咱们来聊聊这作为研究工具的意义，还有临床应用吧。

so so so let's talk about the meaning of this as a research tool and then the clinical implications.

对吧？

Right?

这作为研究工具到底有什么意义呢？

Like, why is it meaningful as a research tool?

这为什么算得上是一个有意义的概念验证呢？

Why is this a meaningful proof of of concept?

首先，它们在实验室里是能起作用的。

The the first thing is that they're they work in the lab.

对吧？

Right?

没错。

Uh-huh.

当你真正创建出这些AI设计的序列时，就能包装出真实的噬菌体颗粒，它们能在实验室里像普通噬菌体一样感染活细菌。

That when you actually create these AI designed sequences, you can actually package real phage particles, and they can actually infect living bacteria in the lab the way that a normal one would.

这算是第一步。

Now that's sort of step one.

说清楚啊，这个病毒是AI凭空想出来的，以前根本不存在的东西。

Just to be clear, this is a a thing, virus a that has never existed before, that that the AI invented.

没错。

Exactly.

它真的就生效了。

It just worked.

机器随便编了个东西，结果真能用，虽然它是病毒，但已经算半条命了。

A machine made it up, and it and it worked, and it was quasi alive because it's a virus.

是的。

Yeah.

对极了。

Exactly.

光是这一点就已经很厉害了。

And so that that alone is pretty cool.

第二点是我们能控制生成过程。

The second thing is that we could steer the generation.

你可以用它专门感染你关心的大肠杆菌菌株，而不影响其他你不关心的。

And so you can use it to infect specific strains of e coli that you cared about and not others that you didn't.

比如说，在噬菌体疗法里，这其实有个很关键的治疗意义，对吧？

And so, you know, in phage therapy, for example, misses a therapeutic implication of this, right?

如果你要针对肠道，那你肯定只想干掉那些有害的肠道菌，而不是把所有菌一锅端——毕竟广谱抗生素就是这么干的，把整个复杂的肠道菌群全给毁了。

If you were to, you know, target the gut, you would want to target specific microbes, for example, that are like bad gut bacteria and not And others that are so you don't want to just do a broad spectrum wiping out of, you know, the entire complex gut community, which is what broad spectrum antibiotics do.

但理论上，你可以更精准地只清除你目标的那部分菌群，对吧？

But you could do, in principle, much more selective deletion of something that you want, right?

从基础科研的角度看，这个模型能精准控制生成，就能在实验室里实现非常精确的功能效果，而正如我刚才说的，这对噬菌体疗法也有非常有意思的治疗潜力。

From a basic science point of view, the controllability of generation from the model can lead to important, precise functional outcomes in the lab, and then this has, I think, as I just explained, really interesting therapeutic implications for phage therapy.

就像精准治疗一样。

-Like a targeted therapy.

就是精准治疗，对吧？

-A targeted therapy, right?

从某种角度说，这两种医学都非常看重精准和选择性，对吧？

And in a way, both medicines well, we care a lot about precision and selectivity, right?

是的。

Yeah.

这其实就是治疗效果和副作用控制的核心所在，对吧？

And that's really the entire game of on target therapeutic efficacy and side effect profiles, right?

没错。

Right.

你希望它只做好一件事，别的什么都别干。

You want it to do one thing really well, and you want it to do nothing else.

你希望它安全。

You want it to be safe.

你希望它，对。

You want it to Yeah.

要

Be

对

Yeah.

还要有效

And effective.

没错

Exactly.

安全又有效

Safe and effective.

能跟我讲讲你和EVO在BRCA1基因上的研究吗？这个基因和某些乳腺癌病例有关。

Tell me about the work you're doing with EVO and the BRCA1 gene, this gene that's implicated in some cases of breast cancer?

你知道，BRCA1是个重要的基因，可能引发乳腺癌或卵巢癌。

You know, BRCA one is an important gene that can cause breast or ovarian cancer.

是的

Yeah.

我们想用这个模型来弄清楚：一个基因突变到底会不会导致疾病。

The thing that we wanted to see with the model is can it understand whether or not a genetic mutation will cause disease?

比如有人来做基因测序，发现自己BRCA1基因有突变，作为女性，她就会问：我是不是患乳腺癌的风险更高了？

And so you can feed in, you know, if someone comes in and gets a, you know, gets their genome sequenced, then they want to know, as a woman, my I have a mutation in the BRCA1 gene.

我是不是得做双侧乳腺切除？还是说，只要每年做一次乳腺X光检查、定期监测就行？

Do I need am I at higher risk of getting breast cancer?

很多患者一旦发现是致病性变异，就选择切掉双乳，但也有人只选择定期检查，对吧？

Do I need to get a double mastectomy if I have a causal variant, many patients elect to do that, or do I just do an annual mammogram and I just monitor, right?

但大多数情况下，你的突变和临床数据库里记录的那些都不一样。

And most of the time, your mutation is different from the ones that are recorded in existing clinical databases.

所以它就被归类为‘意义未明的变异’，对吧？

So it becomes classified as a variant of unknown significance, right?

而这个模型能很准确地判断：这种前所未见的突变，虽然我们还不知道它具体干嘛的，但到底会不会致病。

And the model has a very good sense of whether or not that variant that has never been seen before, you don't know what it does, whether or not it will cause disease or not.

对吧？

Right?

而且它不只是针对BRCA1基因，对任何基因都适用。

And it does that not just for BRCA1, but for any gene.

你怎么能确定呢？

How do you know that?

你怎么验证这种模型在新样本上的表现？

How do you test that sort of out of sample?

其实呢，我们会拿它跟一些权威数据库做对比，比如国立卫生研究院建立的那些金标准数据集。

Well, you you you essentially, you benchmark it, right, against, there are these, like, ground truth databases created by the National Institutes of Health that we've tested against.

另外，你还可以做实验验证。

And then you can also, again, you can run experiments.

比如在实验室里人工制造一些意义不明的变异，然后用实验检测它们的实际影响，再跟已知致病或无害的突变对比，看看它们在致病性评分上排在哪里。

So you can, for example, generate those variants of unknown significance in the lab and actually test them in a lab assay and compare them against the causal mutations or the benign mutations and essentially see where they rank in terms of, you know, this metric of pathogenicity.

对吧？

Right?

已经有人做过这些实验了，这些数据就能当基准数据集用。

And so folks have done those experiments, and you can use those as benchmark data sets.

所以，这个看起来确实有直接的临床应用价值。

So, I mean, that one seems to have kind of immediate clinical relevance.

是这样吗？还是我漏掉了什么？

Does it, or am I missing something?

我们绝对不建议大家去用Evo Designer这个工具，把基因序列贴到Arc网站上，然后凭这个做临床决策。

We certainly don't think that folks should be using, you know, Evo Designer on the Arc website, putting in their genetic sequence and then trying to make clinical decisions.

对吧？

Right?

当然。

Sure.

但我们正在内部训练新一代模型，相信它能在这种基因诊断任务上达到顶尖水平。

But we are making newer versions of the model, which we're training internally, that we think can be state of the art at doing this type of genetic diagnostic.

所以我们很期待能不断优化这些模型。

And so we're excited about improving these models over time.

我觉得，要想让医生在临床诊断中用上这些工具，还得经过大量测试、验证和监管审批，确保它能被正确认证和使用。

I think in order for them to be used by a doctor in the context of a clinical diagnosis, right, there's a lot of testing and validation and regulatory oversight that will need to happen to make sure that this can be used and certified in the right way.

但从研究角度来看，这真的太让人兴奋了，毕竟模型现在的表现已经很出色了，更别说我们还在后台做更多新改进。

But from a research point of view, right, I think this is extremely exciting given the performance of the model already, let alone the new things that we're doing to it under the hood.

我们马上回来。

We'll be back in just a minute.

做生意却不考虑做播客？

Run a business and not thinking about podcasting?

你再想想吧。

Think again.

听播客的美国人，比听Spotify和Pandora广告版流媒体音乐的还多。

More Americans listen to podcasts than ad supported streaming music from Spotify and Pandora.

作为头号播客平台，iHeart的规模是第二名和第三名加起来的两倍。

And as the number one podcaster, iHeart's twice as large as the next two combined.

不管你的客户听什么，你的信息都能传到他们耳朵里。

So whatever your customers listen to, they'll hear your message.

而且，只有iHeart能让你的信息覆盖到广播电台的听众。

Plus, only iHeart can extend your message to audiences across broadcast radio.

你觉得播客能帮到你的生意吗？

Think podcasting can help your business?

那就选iHeart。

Think iHeart.

流媒体、广播，还有播客，全都有。

Streaming, radio, and podcasting.

拨打844844，马上开始。

Call 844844 to get started.

就是844844。

That's 844844

很多职场建议听着挺有道理，可真用起来就崩了。

A lot of work advice sounds good in theory, but falls apart when you actually try to use it.

我就是要改变这种状况。

I'm on a mission to change that.

我是莫莉·格雷厄姆，创业者，也是TED出品的《工作生活》播客的新主持人。

I'm Molly Graham, a company builder and the new host of Work Life, a podcast from TED.

我职业生涯一直待在高速成长的公司里，我最清楚的一点就是：工作从来就不是井井有条的。

I've spent my career inside fast growing companies, and one thing I know for sure is that work is messy.

这一季，我很期待跟大家分享我跟创业者、管理者和创意人之间的对话，聊聊他们最光鲜成就背后的真实故事——那些没人会在领英上发的教训。

In this new season, I'm excited to share my conversations with founders, operators and creatives about the real story behind their shiniest successes, the lessons that no one ever posts on LinkedIn.

现在就去你常用的播客平台收听《Work Life》吧。

Listen now on Work Life wherever you get your podcasts.

之前对话里，帕特里克提到过BRCA1这个基因，它和乳腺癌直接相关。

Earlier in the conversation, Patrick was talking about the BRCA one gene with that single gene linked to breast cancer.

但大多数有遗传因素的疾病，并不是由某一个单一基因决定的。

But for most diseases with genetic components, there is not one single gene.

而是涉及很多基因，而且在很多情况下，我们其实还不太明白到底发生了什么。

There are lots of genes, and in lots of cases, we still don't really understand what's going on.

所以我问他，这类疾病的复杂性，是怎么影响他工作的。

And so I asked him how that universe of diseases fits into his work.

对。

Yeah.

所以这部分挺有意思的，说的就是我。

So this is some of the fun part, me.

咱们到现在聊的都是单个基因里的突变，对吧？

Obviously, you know, so far we've been talking about individual mutations in one gene, right?

对，就是在这种基因明确、疾病也相对清楚的完美条件下，

Yeah, And, you know, in this very kind of well controlled setting in a very well understood gene for a reasonably well understood disease, right?

但生物学的大部分情况都没这么简单。

Most of biology is not that simple.

对吧？

Right?

没错。

Right.

所以我们才需要全基因组关联研究和多基因风险评分，因为很多性状太复杂了，一堆突变共同作用，产生五花八门的表现。

And so the reason why we have GWAS studies or polygenic risk scores is because we have complex traits, lots of different mutations that create lots of different phenotypes.

GWAS，全基因组关联分析。

GWAS genome wide association.

咱们直接看整个基因组，找找哪些和性状有关联

Let's look at the whole genome and see what correlates

没错，就是这样。

with Exactly.

这简直就是一场大规模的人类遗传学钓鱼实验。

It's basically the ultimate human genetics fishing experiment.

到底是什么基因原因，让人有这么个特征？

What the hell is the genetic reason why someone has a given trait?

对吧？

Right?

这个特征可以是任何东西——你找一群人有这个特征A，再找一群看起来正常的人，把他们全测序，看看哪些基因在有特征的人群里更常见，对吧？

And that can be literally any trait, and you get a bunch of people that have that trait A, and a bunch of people that seem normal, sequence them and look at the statistical association for what genes seem to be enriched in the people who have the desired or bad trait, right?

可能是身高，可能是头发粗细，也可能是得阿尔茨海默病的风险。

And that could be height, that could be hair thickness, that could be likelihood to get Alzheimer's disease.

对。

Yeah.

而且大概十年前，也就是人类基因组计划刚完成那阵，大家都对这个方向抱有很大期望。

And it seems like that was one that people were very hopeful about, I don't know, ten years ago or something, in the kind of post human genome project era.

可真做起来才发现，这件事比所有人预想的都要难得多。

That turned out to be a lot harder than anybody thought.

是这样吧？

Is that right?

对的。

Yeah.

我记得当初我们测序人类基因组的初衷，是想挖掘出一大批药物靶点。

I think our initial goal with, sequencing the human genome was that we would find a whole bunch of drug targets.

结果后来我们发现，得先有能力测更多更多人的基因组，说不定还要测所有人的，再用人工智能来处理这些复杂到看不懂的海量数据，才能真正弄明白这些突变到底是怎么互相作用、怎么发挥影响的，你说对吧？

It turned out we needed the ability to sequence many, many more genomes, potentially everybody's genomes, and compute over all of this inscrutable complex data with AI, right, in order to actually understand how these mutations actually interact and work, right?

这也是为什么我们对下一代EVO模型特别期待——我们想用它来研究基因间的相互作用，还有多基因控制的性状，而不是那种只由单个基因、单个突变引发的单基因性状。

And so that's the thing that we're very excited about with next generation versions of EVO is to try to understand genetic interactions and polygenic traits rather than monogenic traits, that are caused by a single gene and a single mutation.

毕竟绝大多数疾病，或者说影响最多普通人的那些病，本来就不是单个基因的问题，是一大堆遗传因素共同作用的结果，而且...

Because most diseases or the diseases that affect most people certainly are have lots of genetic inputs and it's

不只是

not just

一个基因出问题那么简单。

one messed up gene.

对。

Yeah.

你上高中的时候学过，基因型和环境共同作用形成表型，说白了就是先天和后天一起决定了人的行为和所有生物学特征，对吧？

So in high school, you learned that genotype and the environment collaborate to create phenotype, which is a fancy way of saying that nature and nurture collectively create behavior and all like biology, right?

是啊。

Yeah.

所以，这正是我们做的‘邪恶项目’和‘虚拟细胞’研究开始交汇的地方。

And so this is really where our evil work and our virtual cell work starts to collide.

一个基因型的模型，加上细胞反应和行为的模型，我们该怎么把这两个模块真正连起来，才能更准确地模拟细胞生物学呢？

A model of genotype and a model of cellular response and behavior, and how do we actually connect these two modules in order to make more accurate models of cell biology.

对吧？

Right?

你刚才提到了虚拟细胞，咱们还没怎么聊过，这看起来像是Arc的另一个重大项目。

So you you brought up the virtual cell, which we haven't really talked about, which seems like kind of the other, at least another big project at Arc.

对吧？

Right?

给我讲讲这个虚拟细胞吧。

Tell me about the virtual cell.

咱们之前聊过GWAS。

So we talked earlier about GWAS.

对吧？

Right?

GWAS是基础性的研究

And GWAS is fundamental

全基因组关联分析。

Genome wide association.

没错。

Exactly.

全基因组关联分析。

Genome wide association.

这只是一个关联。

It's an association.

对吧？

Right?

这不代表因果关系。

There's no causality.

对吧？

Right?

所以，CRISPR和基因编辑的重大突破，就在于能把那些有关联的突变真正复制出来，在实验室和活细胞里做实验，观察细胞的反应和行为。

And so in many ways, the breakthrough of CRISPR and genome editing was to basically take mutations that are associated, causally creating them and testing them in the lab and living cells and seeing how the cells respond and behave.

如果你制造出这种和疾病相关的突变，正常细胞会癌变吗？

When you make this disease associated mutation, do you get cancer from a normal cell?

这就让我们能说清楚：我们确实发现有相关性，但还不知道是不是因果关系。

So it's allowing us to say, well, we know there's correlation here, but we don't know it's causative.

至少在某些情况下，它能帮我们回答这个问题。

It allows us to answer that question, at least in some context.

没错。

Exactly.

这些实验都是在实验室里做的。

And these are experiments that you do in the lab.

那现在的问题是，我们能把多少这类工作交给AI来做？

Now the question is, how much of this can we push to an AI model?

不用再做CRISPR实验了，模型直接就能告诉你结果。

So instead of doing the CRISPR experiment, the model can just tell you.

这不又回到老问题了嘛——用CRISPR一个一个在实验室里做，实在太慢了？

And is this going back to the original problem of, like, it's wildly slow to have to do these one by one in the lab with CRISPR?

Exactly right.

That's exactly right.

不管是CRISPR、小分子库，还是别的什么方法，都一样。

With CRISPR, with a small molecule library, or anything else.

对吧？

Right?

关键是，我们想要一个能真正做出预测的模型。

The point is we want a model that can actually predict.

比如预测CRISPR实验或者药物处理的结果，对吧？

For example, the results of CRISPR experiments or drug perturbations, right?

没错。

Uh-huh.

还能指导实验室里人们要做的实验。

And guide the things that people are going to do in the lab.

所以意思就是，如果我们取一个细胞，改变某个基因的表达，或者用某种药物处理它——用个不太准确的说法——会发生什么？

So it's basically saying, if we took a cell and we changed the output of this gene, or if we took a cell and we used this drug on it, for lack of a better phrase, what would happen?

明白吗？

Question?

是的。

Yeah.

对我来说，可以这么想：你就像个DJ。

The maybe the mental model for me is imagine you're like a DJ.

对吧？

Right?

你面前摆着地球上最复杂的混音台。

And you have the most complex mixer, on the planet.

对吧？

Right?

你正站在拉斯维加斯一家夜店的中央。

And you're in the middle of this Las Vegas club.

面前全是旋钮和推子。

You have all these knobs and dials.

每动一下，音乐就变，你其实是在调控细胞的‘音乐’，懂吧？

And every time you move them around, the music changes, and you're controlling the music of the cell, right?

当你把某个旋钮拧到最大，某种‘音效’就会变得特别混乱、震耳欲聋，就像疾病爆发一样。

And so when you turn this up all the way, you know, for some given, you know, sound, you can get this kind of chaotic, super loud disease like noise.

听起来一点都不悦耳，也没意思，对吧？

It doesn't sound mellifluous and interesting, right?

然后你就想，怎么才能把这玩意儿关掉。

And then you want to basically figure out how to turn it off.

但如果这噪音其实是好多不同声音混在一起造成的，那你得赶紧找出成千上万个旋钮里，到底哪个该调、怎么调。

But if it's actually a bunch of different sounds that are contributing to the noise, right, you need to figure out the fastest way to find out which of these, you know, you know, tens of thousands of knobs are out there should be actually changed in the right way.

现在我们都是靠瞎猜，猜哪个旋钮该动，然后瞎折腾一通。

Now today, we try to find this out by trying to guess which knobs we need to be dialing, and then we, like, just mess around and do it.

这方法太慢了，完全是组合爆炸，效率低得要命。

And this is extremely slow, and it's this combinatorial search that's incredibly slow.

不是只有一样东西在作怪。

It's not one thing.

也不是说你只要猜中一个关键点就行。

It's not even that you have to guess the one thing.

而是得好多件事同时对了才行。

It's like, it's many different things have to happen.

所以如果你靠试错，这背后的数学复杂度简直离谱，耗时太长了。

And so the math of that, if you're doing trial and error, is just crazy and it takes forever.

对。

Yeah.

这完全是组合爆炸的问题。

It's exclusively combinatorial.

而这个模型的目标是当个智能助手，就像你戴的AR眼镜一样，直接告诉你：‘调这个旋钮、这个、还有这个’，或者‘你该试试这五个旋钮和那五个’，这样你就能更有针对性地去调整了。

And the goal of the model is to be a copilot, where it would be the equivalent of your, like, meta AR glasses that tells you it's this knob and this knob and this knob, or it's you should try these five knobs and these five knobs, and then you can you know in a much more targeted way, you know, which things to dial.

对吧？

Right?

这样一来，我们就能大大加快对模型预测的实验验证，更快找到能治疗疾病细胞的精准扰动方案。

And that's how we could accelerate the experimental verification of the model's predictions in order to try to find the right perturbations that could treat disease cells, for example.

当然，这种能力不仅能用于基础科研，对药物靶点发现也特别关键——这正是我们最想加速的领域。

But this sort of, of course, is a platform capability that could be useful across basic science, but also very much for therapeutic target ID, which is something that we deeply care about accelerating.

是的。

Yeah.

我的意思是，治疗靶点其实就是说，我们应该让药物去修复哪个目标

I mean, therapeutic target idea is basically like, what should we send a drug to go fix

没错。

Exactly.

或者调整一下。

Or change.

对吧？

Right?

没错。

Exactly.

我记得你之前提到过，临床试验的失败率特别高。

I mean, you've talked about I think in this context, I've heard you talk about the the failure rate of clinical trials.

是的。

Mhmm.

对吧？

Right?