Pod Insights a16z Podcast a16z Podcast: Deep Learning for the Life Sciences About this episode with Vijay Pande (@vijaypande) and Bharath Ramsundar
Deep learning has arrived in the life sciences: every week, it seems, a new published study comes out... with code on top. In this episode, a16z General Partner Vijay Pande and Bharath Ramsundar t...
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Hanne Tidnam? Hi and welcome to the A16Z Podcast. I'm Hannah. Deep learning has come to the life sciences. Lately, it seems, every week, a published study comes out with code on top. In this episode, A16Z general partner on the biofund, Vijay Pandey and Bart Rumsundar, talk about how AI and ML is unlocking the field in a new way, in a conversation around their recently published book, Deep Learning for the Life Sciences, written along with co-authors Peter Eastman and Patrick Walters. The book aims to give developers and scientists a toolkit on how to use deep learning for genomics, chemistry, biophysics, microscopy, medical analysis, and other areas. So, why now? What is it about ML's development that is allowing it to finally make an impact in this field? And what is the practical toolkit, the right problems to attack, the
0:44 Hanne Tidnam? right questions to ask? Above and beyond that, as this deep learning toolkit becomes more and more accessible, biology is becoming democratized through ML. So, how is the hacker ethos coming to the world of biology? And what might open source biology truly look like?
1:00 Vijay Pande? So Bart, we spend a lot of time thinking about deep learning in life sciences. It's a great time, I think, for people to become practitioners in this space, especially for people maybe that's never done machine learning before from the life sciences side, or maybe people from the machine learning side that get into life sciences. But maybe even the place to kick it off is, you know, what's special about now? Like, why should people be thinking about this?
1:23 Speaker 02 The challenge of programming biology has been that we don't know biology, and we make up theoretical models, and the computers are wrong, and biologists and chemists understandably get grumpy and say, why are you wasting my time? But with machine learning, the advantage is that we can actually learn from the raw data. And all of a sudden, we have this powerful new tool there. It can find things that we didn't know before. And this is why now's the time to get into it, really to enable that next wave of breakthroughs in the core science.
1:51 Vijay Pande? The part that still blows me away is just how fast this field is moving. And it feels like it's a combination of, you know, having the open source code on places like GitHub and Archive, and there's like a paper a week that's impactful, when it used to be maybe a paper a quarter or a paper a year. And the fact that code is coming with a paper, it's just layering on top. I mean, that seems to me to be sort of the critical thing that's different now.
2:16 Speaker 02 I think when you can clone a repo off GitHub, you all of a sudden have new insights just because I'm using a new language. And now that thousands of people are getting into it, I think all of a sudden you'll find lots of semi-self taught biologists who are really starting to find new, interesting things. And that is why it's exciting. It's like the hacker ethos, but kind of coming into the bio world, which has typically been much more buttoned down. Now, I think anyone who can clone a repo can start really making a difference. I think that's going to be where the real long-term impact arises from these types of efforts.
2:48 Vijay Pande? You don't need a journal subscription to get archive or to get the code, which is actually, that alone is kind of amazing. It wasn't that long ago where a lot of academic software was sold, you know, and it may be sold for $500, which is very material. That's one piece. You connect that to the concept of now AI or ML can unlock things in biology. Then biology is becoming democratized is kind of sort of your point, right? And so let's talk about that because we're still learning biology collectively. What is it about deep learning in biology now? Because biology is old, machine learning is old. Like, what's new now?
3:23 Speaker 02 Deep learning has this question all over the place. Why does it work now? The first neural nets kind of popped out in the 1950s. And I think it's really a combination of things. I think that part of it is the hardware, really. The hardware, the software, the growth of kind of rapid linear algebra stacks that have made it accessible. I think also an underappreciated part of it is, you know, the growth of the cloud and the internet, really. You know, neural nets are about as janky now as they used to be in the 80s. The difference is that I can now pull up a blog post where someone says, oh, these things are janky. Here's the 17 things I did. I can copy paste that into my code and all of a sudden I'm a neural net expert. It's not quite that easy. It turns into a tradecraft almost that you can learn by just working through it.
4:06 Speaker 02 That's why the deep learning toolkit has been accessible. Then you get to biology and the question is, why biology? Why now? And I think here actually the question's a little deeper. I think that. It's really about, I think, representation learning. So we have now reached this point where I think we can learn representations of molecules that are useful. This has been something that, you know, in the science of chemistry, we've been doing a long time. There's been all sorts of, you know, hand encoded representations of parts of molecular behavior that we think are important. But I think now, using the new technology from, you know, image processing, from word processing, we can begin to learn molecular representations. And, you know, to be fair, I actually don't think we've really broken through there.
4:50 Speaker 02 If you look at, you know, what's happening in images or text, they're five years ahead of us.
4:55 Vijay Pande? Well, let me break in here because just for the listeners to give a sense for why representation is important, one of my pet examples is that if I gave anybody like, say, two five-digit numbers to add, it'd be trivial. If I gave you those same five-digit numbers in Roman numerals and you want to add them, the representation there would make this like insane. And what would you do? Well, you would convert into appropriate representation where the operations are trivial or obvious. And then the operation is done and maybe re-encode, auto-encode back to the other representation. So this is the problem is like when you have a picture, representations are obvious because it's pixels, right? And computers love pixels. And maybe even for DNA, like DNA is like a one-dimensional image and so you have bases that are kind of like pixels.
5:40 Vijay Pande? We used to joke early days that we would just take a photograph of the small molecule and then use all the other stuff, but that's kind of insane too. And so with the right representation, things become transparent and obvious, with the wrong representation becomes hard.
5:52 Speaker 02 This is really at the heart of machine learning. It's that there's something about the world that I want to compute on, but computers only accept, you know, very limited forms of input, zero ones, tax strings, like simple structures. Whereas like if you take a molecule, a molecule is like a frighteningly complex entity. So one thing that we often don't realize is that Until 100 years ago, like, we barely had any idea what a molecule was. It's this alarmingly strange concept that although we see little, you know, diagrams in 10th grade chemistry or whatever, that isn't what a molecule is. It's a much weirder, weirder quantum object, dynamic, kind of shifting, flowing. We barely understand it even now. So then you really start asking the question of like, what is water?
6:36 Speaker 02 For example, is it three characters, H2O? Is it, you know, two hydrogens, an oxygen? Is it some quantum construct? Isn't like this dynamic vibrating thing? Is it this bulk mass? There's so many layers to kind of the science of it. So what you really want to do is like, you got to pick one. And this is where it gets really hard, right? Like, if I'm thirsty, what I care about in water is a glass of water.
7:00 Vijay Pande? Yeah.
7:01 Speaker 02 If I'm trying to answer deep questions about, you know, the structure of Neptune, I might want a slightly different representation of water. The power of the new deep learning techniques is we don't necessarily have to pick a representation. We don't have to say water is X or water is Y. Instead, you say, let's do some math and let's take that math and let the machine really learn the form of water that it needs to answer the question at hand. So one form of mathematical construct is, you know, thinking of a molecule as a graph. And if you do this, you can begin to do these graph deep learning algorithms that can really extract meaningful structure from the molecule itself. We've learned finally that here's a general enough mathematical form we can use to extract
7:46 Speaker 02 meaningful insights about molecules or these critical biological, chemical entities that we can then use to answer real questions in the real world.
7:55 Vijay Pande? What I think is interesting here in particular is that so much has been developed on images and there's a lot of biology that's images. And so we could just spend the whole time talking about images and it could be microscopy or radiology and tons of good stuff there. But there's a lot of biology that's more than images and molecules is a good example. And, you know, for a long time, it seemed like deep learning was being so successful in images that that's all it really did. And if you could sort of take your square peg and put in whatever holes you got, it would work. What you're talking about for graphs is kind of an interesting evolution of this because a graph and an image are different types of representations. But, you know, at a technical level, convolutional networks for images or graph convolutions for graphs are kind of a sort of borrowing a concept at a higher level.
8:39 Vijay Pande? The biology version of machine learning is starting to sort of grow up and starting to not just be a direct copy of what was done with images and in other areas, but now starting to be its own thing.
8:50 Speaker 02 A five-year-old can really point out the critical points in an image, but you almost need a PhD to understand the critical points of a protein. So you have this like dual kind of weights of burden of understanding. So it's taken a while for the biological machine learning approach to really mature because we've had to spend so much time even figuring out the basics. But now we're finally at this point where it feels like we are diverging a little bit from the core trunk of, you know, what people have done for images or text. In another five years, I'm going to be blown away by what this thing does. It's going to understand more deeply.
9:28 Vijay Pande? So we kind of have this sort of connection between democratization of ML, ML into biology, democratization into biology. But I don't think we're there yet. I think for ML, I think there really is a sense of democratization. You could code on your phone and do some useful things, or certainly on a laptop, a cheap laptop. I mean, but for biology, what is missing? One is data, and there's a fair bit of data. In the book, we talk about the PDB, we talk about other data sets, and there are publicly available data sets. But somehow that doesn't get you into the big leagues, you know? So like if in this vision of democratizing biology, what's left to be done?
10:06 Speaker 02 In some ways, the democratization of ML is a teensy bit of an illusion even. It's because that the core constructs were mathematically invented, that there is this convolutional neural net, or its cousins, the LSTM, or, you know, the other forms of core mathematical breakthroughs that have been designed, that you can take these building blocks and just apply them straight out. In biology, as you pointed out earlier, I think we don't have those core building blocks just yet. We don't know what the Lego pieces are that would enable a newcomer to really start to do, you know, breakthrough work. We're closer than we were. I think we have the beginnings of a toolbox, but we're not there yet.
10:49 Vijay Pande? Well, let's think about what happened on the ML side as inspiration for the bio side. How much is it driven through academia? How much is driven through companies? Because what I'm getting at is that there's a lot of bio in academia. I don't know if we're seeing that being made open source in companies.
11:02 Speaker 02 We're getting to this really weird set of influences where In order for companies to gain influence, they need to open source. So this is why, you know, 10 years ago, I can't imagine that Google would have open sourced TensorFlow. It would have been core proprietary technology. But now they know that if they don't do that, developers will shift to some other platform.
11:23 Vijay Pande? Right, by some other company.
11:24 Speaker 02 Exactly. So it's weird that the competitive market forces are driving democratization.
11:30 Vijay Pande? Well, so PyTorch, basically sort of Facebook based, right? And TensorFlow is from Google. Let's say Google kept TensorFlow proprietary. What would be so bad for them if they did that? What if everybody outside used PyTorch?
11:41 Speaker 02 This actually, I think there's like a really neat kind of analogy to the financial sector. A lot of financial banks have masses of functional programs that they keep under the hood, under the covers. If you look at Jane Street, or I believe Standard and Chartered, or a few other of these other big institutions. Lots and lots of functional code hiding behind those walls. But that really hasn't really infiltrated further out. And this actually, I think, in the long run, weakens them because it's harder to train. It's harder to find new talent. It's more specialized. A lot of the code base at Google is proprietary. Like, the original MapReduce was never put out there. And this, I think, has actually caused them a little bit of a problem
12:21 Speaker 02 in that new developers coming in have to spend months and months and months getting up to speed with the Google stack. Whereas if you look at TensorFlow, it doesn't take any time at all. Someone could walk in and basically be able to write.
12:34 Vijay Pande? They've been using it for months to years.
12:36 Speaker 02 Exactly. And I think at the scale that big tech is at, this is just like, it's a powerful market advantage. They're almost outsourcing their education process.
12:44 Vijay Pande? Yeah, and I guess if they don't put it out, someone else will.
12:47 Speaker 02 Exactly.
12:47 Vijay Pande? And then they'll learn on their platform. Yes, but then maybe what is the missing part in biology? We've got pharma, you know, a huge force there, but they have very specific goals. A lot of agricultural companies, but it's much more disparate.
13:01 Speaker 02 Yeah, it's dramatically hard to actually take an existing organization and turn it into an AI machine learning organization. And so one thing. I've honestly been surprised by is that, you know, when I've seen companies or organizations I know try to incorporate AI into their drug discovery process, it ends up taking them years and years and years because they're fighting all these upstream battles. Weeks to get their, you know, old computing system to upgrade to the right version of, you know, their numerical library so they could even install TensorFlow. And then they had all these things about who can actually say, upgrade kind of the core software, who is it kind of this department, like, how much does it need to talk to the
13:45 Speaker 02 biologist, to the chemist? The fact is that pharma and, you know, existing big co's are not built this way. That's not their core expertise. Whereas if you look at, you know, Facebook or Google, they've been machine learning for almost two decades now, like, from the first AdWords model. So in some sense, like, they had to change very little about their culture. Like, yeah, there's a slight difference instead of like this function, use that function, but whatever. But the core culture was there. Exactly. And I think the culture, the people, changing that is going to be dramatically hard. So which is why I think it will really take, I think, ten years and a generation of students who have been trained in the new way to come in and shift.
14:24 Vijay Pande? Yeah, well, Google was a startup too, right? I think, you know, the thesis was that, and is that startups will be able to build a new culture. And I think the key thing that I think we're seeing sort of boots on the ground is that that culture has to be not, here's your data scientists or machine learning people in one room and your biologist in another room, that they'd have to be the same. What's intriguing to me is just the size of the biomarket. Biology is healthcare, it's agriculture, it's food, it could be the future of manufacturing. There are so many different places that biology plays a role today and will play a role. But it just means that I think, I think to the point we're talking about, these companies just are being built right now.
15:04 Speaker 02 There's, I think, this whole host of challenges here because biology is hard.
15:10 Speaker 02 And building kind of like that selective understanding of like, you know, of the 10 best practices that existed, five are actually still best practices. The other five, we need to toss out a window and stick in a deep learning model. That kind of very painstaking process of experimentation and understanding. That I think is like where the really hard innovation is happening. And that's going to take time. You're never going to be able to replace like a world-class biologist with any machine learning program. A world-class biologist is typically fricking brilliant. And they often bring a set of understanding that no programmer or no computer scientist can. Now, the flip side holds true. And I think that merger, as you said, that's where like there's power for magic dynamism.
15:53 Speaker 02 One really interesting factoid I heard from an entrepreneur in this space is that, you know, the best biologist that they could hire had a market rate that was lower than a introductory intermediate, you know, front-end dev.
16:07 Vijay Pande? Yep.
16:08 Speaker 02 And, you know, of course, front-end is very hard engineering. Don't want to put that down. But there's so many fewer of these biologists. So there's almost this market imbalance of how is it possible that, you know, you can take really a world-class biologist of whom there's maybe a couple hundred in the world and not have them be valued properly by the market? So do you even out those pay scales in one company? Do you like have two awkward pay ladders that coexist and create tension in your company? These are the types of like really hard operational questions that almost have nothing to do with science, but at the heart of it, they do.
16:43 Vijay Pande? Maybe it's interesting to talk about like how we can help people get there.
16:45 Speaker 02 Mm-hmm.
16:46 Vijay Pande? You know, so what's like the, the training they should be doing? Maybe we could even go like super nuts and bolts. So I got my laptop. What do I do? So I mean, like, I guess there's a couple key packages we install. It's like TensorFlow, maybe DeepGAM, something like that. Python is often already installed, let's say on a Mac. Is that it? And then we start going through papers and books and code?
17:06 Speaker 02 I think the first place really is to You need to form an understanding of like, what are the problems even that you can think about? I think if you're not trained as a biologist, and even if you are, you might not see that intersection of these are the problems where biological machine learning can or cannot work. And that I think is really what the book tries to teach you, as in like, what's the frame of thinking? What's the lens at which you look at this world and say that, oh, that is data coming out of a microscope. I should spend thirty minutes spin up a combat and iterate on that. This is a really gnarly thing about how I prepare my like, you know, C Elegance samples. I don't think the deep learning is going to help me here.
17:49 Speaker 02 And I think it's that blend of knowledge that the book tries to give you. It's like a guidebook. When you see a new problem, you ask, is this a machine learning problem? If so, let me use these muscles. If it's not a machine learning problem, well, I know that I need to talk to someone who does know these things. And that's what we try to give. Vanderring has a great rule of thumb. If, you know, a human can do it in a second, deep learning can probably figure it out. So start with something like, say, microscopy. You have an image coming in and an expert can probably eyeball and say, interesting, not interesting. So there's this binary choice.
18:23 Vijay Pande? Right.
18:24 Speaker 02 And there's some arcane black box that was trained within the expert's head and experience. That's actually the sort of thing machine learning is like made to solve. So really ask yourself, like, when you see something like that, is there some type of perceptual input coming in? Image, sound, text, and increasingly molecules, a weird new form of perception, almost magnetic or quantum. But you have perceptual input coming in. And is there a simple right, wrong, left, right, you know, intensity type answer that you want from it? If you do, that's really a machine learning problem at its heart.
19:00 Vijay Pande? Well, so that's one type of machine learning. And I think the benefit there of that, um, what a human can do in a second, deep learning can do. Especially since in principle on the cloud, you could spin up 100, 1,000, 10,000 servers. Suddenly you've got 10,000 people working to solve a problem and then they go back to something else. That's just something you can't do with people. Or you got 10,000 people working 24/7 as necessary. Can't do that with people. But there's another type of machine learning, which is to do things people can't. Or maybe more specifically, do things individual people can't, but maybe crowds could. You know, so like we see this in radiology, right? Where the machine learning can have accuracies greater than any individual. Kin to what, let's say, the consensus would be, which would be the gold standard.
19:42 Vijay Pande? That's maybe the real exciting part, sort of the so-called superhuman intelligence. Where are the boundaries of possibilities there?
19:49 Speaker 02 One of the biggest problems really with deep learning is that you have some like strange and crazy prediction. Now, I think that there's a fallacy that people fall into of trusting the machine too easily. Because 90% of the time, that's going to be garbage. And I think that really, kind of the challenge of picking out these bits of superhuman insight is to know how to shave off the trash predictions.
20:16 Vijay Pande? Yeah, is 90% an exaggeration or is it really 90%?
20:19 Speaker 02 I like nice round numbers, so that might have just been something I picked out. But there's like this great example, I think, in medicine. So there's scans coming in and the deep learning algorithm was doing like amazing at predicting it. And then like they dug into it and it turned out that the scans came from three centers. One of them had like some type of center label that was like the trauma center or something. There's the other non-trauma center. The deep learning algorithm had, like a kindergartner told to do this, learn to identify the trauma flag and flag those and uptick those. So if you did this like naive statistics of blending them all out, you'd look amazing. But really it's looking for a sticker.
20:58 Vijay Pande? Yeah, I mean, there's tons of examples like that. One with the pathologist with the ruler in there and it's becoming a ruler detector and so on. Like, you know, this AUC, like a sense of accuracy of close to 1.0. We all got to be very suspicious of that because just running a second experiment wouldn't predict the first experiment with that type of accuracy. Anything that's too good to be true probably is.
21:18 Speaker 02 Yeah, I think then you get into the really subtle challenges, which is that, you know, the algorithm tells me this molecule should be non-toxic to a human and should have effect on this, you know, indication. Do I trust it? Is it possible that there is a false pattern there? Humans make these types of mistakes all the time, right? Like, if you have any type of, like, actual biotech, you know that there's going to be molecules made or tCS that are disproven. So you're getting into the hardcore of learning, which is that, is this real? The reality is, we don't have answers to these. We're really kind of trending into the edge of machine learning today, which is that, is this a causal mechanism? Does A cause B? Is it a spurious correlation? And now we're getting to that place
22:02 Speaker 02 where humans aren't necessarily better. We talk about some techniques for interpreting, for looking at kind of what informed the decision of the deep learning algorithm. And we do provide a few kind of tips and tricks to start thinking about it. But the reality is, that's kind of the hard part of machine learning. It's the edge.
22:19 Vijay Pande? The interpreting chapter is one of my favorite ones because it's often sort of become so-called common wisdom that machine learning is a black box. But in fact, it doesn't have to be, and there's lots of things to do, and we are quite prescriptive there. So the interpretability, I think, also is frankly, what's going to make human beings more at peace with this. And this isn't anything unique to machine learning. Like, if you had some guru who's like just spouting off stuff and said, you know, buy this stock X and short the stock Y, and put all your life savings into it, you probably would be thinking, okay, well, maybe, but why? So I think this is just human nature, and there's no reason why our interaction with machines would be any different. What I think is interesting is human beings are notoriously bad at causality.
23:03 Vijay Pande? Like, we kind of attribute things to be causal when they're not causal at all. We do that in our lives from like, you know, why did that person give me that cup of coffee? To why did that drug fail? Like, all these different reasons. There's two big misconceptions about machine learning. One is lack of interpretability. The second one is correlation doesn't mean causation, which is true, but somehow people take that to mean it's impossible to compute causality.
23:25 Speaker 02 Yep.
23:25 Vijay Pande? And that's the part that I think people have to really be educated on because there are now numerous theories of causality. And you could use probabilistic generative models, PGMs. There's lots of ways to go after causality. The whole trick, though, is you need time series data. What's beautiful about biology, or at least in healthcare, is that we've got time series data in many cases. So now perhaps finally, there's the ability to really understand causality in a way that human beings couldn't because we're so bad at it, and machines are good at it, and we've got the data.
23:56 Speaker 02 Can you think of like a place where, you know, in your experience, the algorithms have succeeded in teasing out a causal structure that people missed?
24:03 Vijay Pande? Yeah, so you know, I think in healthcare, we always think about what is leading to various changes, like, you know, this drug having adverse effects, this diet having positive or negative effects. All of these things are being understood and in the sort of the category of real world evidence, which is a big deal in pharma these days. And if you think about it, like a clinical trial is really a poor man's surrogate for not understanding causality. Because if we don't understand causality, you've got to do this thing where it's double blind, we start from scratch, and I'm following it in time, and we see it. If you understood causality, you might be able to just get a lot of results from just mining the data itself. As a great example, you can't do clinical trials for all pairs
24:46 Vijay Pande? of drugs. I mean, just doing it for a single drug is ridiculously expensive and important, but all pairs of drugs would never happen, but people take pairs of drugs all the time, you know? And so their adverse effects from real world data is probably the only way to do it. And we can actually get causality, and there's tons of interesting journal medicine papers sort of saying, aha, we found this from doing data analyses. I think that's just starting out.
25:09 Speaker 02 Honestly, I think that AI, bio AI, drug discovery needs to take a page from the self-driving car companies. In the neighboring self-driving car world, simulators are all the rage. And really because it's that same notion of like causality, almost. Like, there's this structure to the world, like pedestrians walk out, chickens, alligators, whatever, crazy things.
25:29 Vijay Pande? Alligators?
25:30 Speaker 02 I saw this Florida picture. It happens.
25:32 Vijay Pande? Yeah, it happens.
25:34 Speaker 02 So I think there, they've built this amazing infrastructure of being able to run these repeated experiments, almost randomized clinical trials, but informed by real data. We don't yet have that infrastructure in bio world. And I know there's a couple of exciting startups who are starting to kind of move towards that direction. But I think it's when we can really probe the causality at scale. And then, in addition to just probing it, when the simulator is wrong, use the new data point that came in and have the simulator, you know? Learn to fix itself. That's when you get to this really amazing feedback loop that could really revolutionize biology.
26:09 Vijay Pande? Yeah, so you know, we talked about some basic nuts and bolts about how to get started. And the framing of questions, which is a key part. So let's say people, they're set up, they've got their question. Where do they go from there? I mean, in a sense, we're talking about something closer to open source biology. And to the extent that biology is programmable and you know, synthetic biology is, I think, very much, you know, it's been around for a while, but I think it's really starting to crest. How do these pieces come together such that we could finally get to this sort of open source biology, democratization of biology?
26:41 Speaker 02 A big part of this is really the growth of community. There are people behind kind of all these, you know, GitHub pages that you see. There's real decentralized powerful organizations that if you look at the Linux Foundation, if you look at, say, the Bitcoin Core Foundation, there are networks of open source contributors really that form this like brain trust. It's very diffuse. It's not centralized in the Stanford Harvard Med Department or whatever. And I think what we're going to see is the advent of similar decentralized brain trusts in the bio world. As in a network of experts who are kind of spread across the world and who kind of contribute through these code patches. And that I think is not at all new to the software world. We've seen that for decades. It's totally new to biology. It's alien.
27:26 Speaker 02 Like, you'd be surprised how much skepticism there can be at the idea that a non, you know, Harvard trained, say, biologist can come up with a deep insight. We know that to be a fact, right? There is multiple PhDs worth of work in just like the Linux kernel that that community really doesn't care to get that stamp of approval. So I think we're going to see the similar parallel kind of knowledge base that grows organically. But it takes time because you're talking about the building of another kind of almost educational structure, which is this new and exciting direction.
27:57 Vijay Pande? Here's the challenge I worry about the most, which is that like, so you're building a Linux kernel. You can test whether it works or doesn't work relatively easily. Even as it is, there's this huge reproducibility crisis in biology. So how does one sort of become immune from that or at least not tainted by that? Like, how do you know what to trust?
28:15 Speaker 02 And this is a really, really interesting question. And this is kind of shading a little bit almost into the crypto world, right? Like, you could potentially think about this experiment where you have like a molecule. You don't know what's going to happen to it. But maybe you create a prediction market that talks about the future of this molecule. And you could then begin to create these historical records of predictions. And we all know there are kind of like expert drug pickers at Big Pharma who can like eyeball and say that is going through, that is failing. And five years later, you're like, well, shit, okay, guess I was right. There is the beginnings of infrastructure for these feedback mechanisms, but it's a really hard problem.
28:53 Vijay Pande? Yeah, I'm trying to think though, what that would be like. The acute thing is like, you could imagine if it was a simple question like, is this drug soluble? Someone might run a cheap software calculation, someone might do the experiments. And there's different levels of of cost of of having different levels of certainty. You're essentially describing a decentralized research facility. Maybe the question is, who would use it?
29:16 Speaker 02 This is, I think, the really hard part because I think that biopharma tends to be a little more risk averse for good reasons than many other industries. But I actually think that in the long run, this could be really interesting. Because if you have multiple assets in a company, you could kind of like disbundle the assets. And then you could start to get this much more granular understanding of like, what assets actually do well, what assets don't. And if you make it okay for people to like, place a bet on these assets, all of a sudden it's de-risked. Because if you're a big pharma and you're like, I don't really believe that Alzheimer's, you know, molecule does what is claimed, but I want to say like, 15% odds it goes
30:00 Speaker 02 through, I'll just invest 15% of what I would have in another world.
30:04 Vijay Pande? Well, the trick is, and especially what we're talking about now is the world of financial instruments as well, is the trick is you have to be able to know how to risk an asset. And so it could be in the end, one of the first interesting applications of deep learning, machine learning, is to use all the available data to give the ML, maximum likelihood estimator of what we think this asset is going to be. It prices the asset and then people can go from there. It's kind of a fun world where we're sort of thinking about how the financial world, machine learning world, and biology come together to kind of decentralize it and democratize it.
30:34 Speaker 02 I think there's opportunities to kind of like. Allow for more risks, the long tail to be played out. You don't have as many interesting hypotheses that grow dead in the water. Because it wasn't, you know, de-risked enough for a big bet.
30:48 Vijay Pande? Yeah, so you know what, I think the big takeaway for me here is that there is that possible world. But I forget if this is the way you learned how to program. The way many of us did is I learned when I was like 11 on like a, actually a TI 99A. And I was just playing around with it and I learned so much because it was, I could just get my hands right in it. And I think kind of my hope for the book is that it's kind of the equivalent in biology that people can get their hands in it. And I don't know where they're going to go with it. It'd be super cool if they go where we're describing. That's one of the possible any futures, but I think that's what we're hopefully be able to give people.
31:23 Speaker 02 We are opening out the sandbox. Here's what we've learned in kind of these very exclusive academic institutions. Let's throw the gate open, say, here's as much as we know as we can try to distill it down. And do what you will with it, like open source means no permission. So go to town and hopefully do something good for the world is kind of the dream.
31:44 Vijay Pande? That sounds fantastic. Well, thank you so much for joining us.
31:46 Speaker 02 Thank you for having me.
Previous a16z Podcast: The Power of Restorative Justice Next What's Next for Esports Startups Page updated September 30, 2026 · transcript and summary are machine-generated
