a16z Podcast - a16z Podcast: Move Fast But Don't Break Things (When It Comes to Computational Biology)

Episode Date: June 14, 2016

The mindset of "move fast and break things", while great for code, isn't exactly great for the human body. So adding computation to biology -- especially in the slow-moving pharmaceutical industry, where drug approval can take years -- brings with it both opportunities (like drastically faster discovery and assessment) and challenges (the need for hard evidence, not just soft-ware). But there's more: We don't want just better outcomes for healthcare. We want better outcomes at a cheaper price. And that's where machine learning comes in. The benefits of such computation -- i.e., software -- can provide a powerful, frictionless, and far more cost-effective tool for biopharmaceutical research ... but it requires data. So who provides that data? Is it the pharmaceutical companies, or the payers (insurance)? How are organizations incented to overcome intellectual property silos in sharing their data? Especially since it was only relatively recently, Jeff Kindler (the former CEO of the world's largest pharmaceutical company, Pfizer) reminds us in this episode of the a16z Podcast, that the FDA even allowed data to be put in computers vs. on paper. But there's a reason the self-driving car was pushed out of the software and not the auto industry, argues TwoXAR co-founder and CEO Andrew Radin -- and it has to do with the unique nature of the developer's mindset applied to novel problems. The deterministic nature of Moore's Law -- it's not a matter of if, but when -- plays a role too, observes a16z bio fund general partner Vijay Pande. There are things that big data and simulations will be able to accomplish that a hundred lab experiments on animals can't. Still, the two mindsets will have to merge, so we can move fast ... but without compromising quality, safety, and reliability. That's the big difference between computer science and biology after all. image: mattza/ Flickr

The content here is for informational purposes only, should not be taken as legal business tax or investment advice or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund. For more details, please see A16Z.com slash disclosures. Welcome to the A16Z podcast. I'm Michael Copeland. And we are here in the room with three people to talk about computational biology and how compute meets biology and And generally speaking, how health care can get better through means of technology. And to help us do that, we have A16Z's Vijay Ponday. We have Jeff Kindler, who is the former CEO of a little company you might have heard of Pfizer,

and who's now an advisor and investor in the biotech space. And finally, Andrew Raiden, who's the CEO of Tuzar, which is a portfolio company of ours, which is in the computational biology space, not surprisingly. So welcome, guys. It's great to be here. We have seen technology and software in particular, you know, from Marvin. manage point seep into all kinds of industries, right? So we're seeing it in finance. We're seeing it in driverless cars. And what there's been this promise in the healthcare space that technology

will come in and revamp all these kind of very expensive, very time-consuming processes to help us get to better health in the end, better therapeutics, better, you know, testing for that matter. Maybe, Jeff, let's start with you. Why is this problem so hard? And how you've been in this space for a long time and kind of seen it from all sides, where are we headed now and, you know, how do we get there? Sure. Well, first, if I could just go back and give a little analogy to another industry that I think is appropriate. So if you think of Hollywood, the movie business, in the 1930s and 40s, the studios had everybody as an employee of the studio,

the crew, the cast, the writers, the directors, and they just cranked out movies. Today, the movie business is one in which the studios act as syndicators, financiers, financial-oriented organizations, and they put together every movie as a package with people that don't necessarily work for them. I think pharma is in the process of making a similar shift. For many years, until even relatively recently, they were one of the few fully integrated industries in the sense that they did everything in-house from discovery through the end of a product's life cycle.

Over the years, starting with clinical research organizations, they were at the lead of this, they have come to realize that some of the things that they think of as cost centers are better done by people on the outside who are doing it for profit and therefore probably do it better, more efficiently, and so forth. I think to answer your question directly, Michael, why hasn't it happened sooner? I think to a large extent, it's because the farm industry has been so profitable and so successful and is so has been and will always be so concerned with control and quality and the rest that they'd probably come reluctantly to outsourcing over the years because it just wasn't

consistent with the way they did things. But what's happened in the last few years is they've come under greater earnings pressure. They're really looking at their balance sheet and their income statement and they're coming to realize that there's a lot of costs in their system that can be driven out and services provide more appropriately and efficiently by people on the outside. Andrew, that gets right to you. I mean, you are. one of those folks. And here's what's interesting, too, is that you come to this not from a medical background or chemistry for that matter.

So, yeah, just to kind of add a little more color to that comment, our company owns a refrigerator and microwave. That is what we own. And I'm guessing you don't actually use that for anything medical other than... No, yeah, that's where lunch goes, right? And so we acquire resources on demand. And so this ability to go out and get millions, if not tens of millions of dollars, of infrastructure, whether that's computer, whether that's lab resources, use it in that

moment, and that moment might be 10 minutes worth of computation, and then release that resource and don't have to pay for it again. That enables startups and small companies to do things, to have all the types of resources that the big guys have without having to have the huge bank account. So in some sense, it's this unbundling of pharma, but it's also this AWS kind of like, look, we're just going to take the services that we need when we need them and pay for that and that's it. Exactly, yeah. Yeah, I think I just would add to that. I think that we're at a great moment in time because I think the technology, and VJ can speak better to this than me,

but the technology and the capability from the cloud is so much greater than it was, and that comes right at a moment in time when pharma can use it best. There's a tremendous need in pharma to have a greater standardization and automation of its processes. And historically, you know, they're doing some of the things that they do in the same way they did decades ago. So this is a very propitious moment to have technology meet the pharmaceutical business and help them both. As Jeff pointed out, pharma hasn't had to really change because the financial pressure certainly wasn't there. And there may be that pressure now. But reeling things back.

So, for example, the cloud and the banking industry, the bank, the banking, Indian industry, it's like, whoa, whoa, whoa, why would we ever do that? Turns out now the banking industry is all up in there in the cloud. What will it take for Farmer to get there in the same way? Yeah, you know, there's a couple of interesting aspects here. First off is that this isn't something that has to happen immediately, that what will happen is when we'll see gradual and gradual penetration of these services in, first in pilot projects than larger and larger.

And we already see this with compute and Farma, that Farmer is using Amazon. at first there was maybe concerns about security, other things, and what I've heard confidentially from people is that AWS actually has better security than often the way a lot of people would set up themselves. That's the benefit of them being an expert in cloud computing, that they can put the resources not just to have something cheaper but better. And on the cloud biology side,

where we're not talking about compute, but we're talking about real-life biology in the cloud, you could imagine the same type of thing, that it could be done not just cheaper and not just lower overhead, actually done better. And I think when something's cheaper and better, then it's actually a very strong value proposition. And something being very new will mean that it won't have immediate adoption, but as people start to see it, I think they'll start to want more and more of it.

So I'm involved with a company that's still in stealth, but you'll hear a lot about it very soon. And what it's doing is addressing a sort of small corner, not much spoken about part of the pharmaceutical industry, which is animal models. Animal models today are done. pretty much the way they were done the 1950s. They're very high touch. They involve a lot of human subjectivity, inaccuracy, unreliability. It still involves people going in and measuring the size of a rat's foot and that kind of thing. And this company has digitized that with a heavy dependence on ADWS, as a matter of fact. They have automated it. They have made it more reliable. They're using sensors and cameras and big data to make that whole system both, as VJ says, both much less

expensive and much more reliable. That's a good example of a part of the pharma industry that was historically done either in-house or by somewhat commoditized outside providers and really no technological innovation for years. And I think that's a precursor of a lot of things to come. I want to get into why there's pressure in pharma, but Vijay, you mentioned cloud biology and Andrew. So those two words don't seem to go together. And just in the way, Jeff, that you describe this, kind of like there's actual real animals, but somehow technology interfaces with them. What is cloud biology? Where are we today

with it? And sort of where do we head? There are several companies in this space that are pushing the envelope of cloud biology. And when I think about it, I think about it as doing real-life experiments. So I don't want to confuse this with doing simulations or calculations. These are real-life experiments, whether that be animal models with mice or in vitro experiments. But the difference here is that it's something that's completely systematized. For example, a lot of in vitro experiments can be driven by robots. And what's intriguing here is that when you drive something by robots, biology now doesn't look like humans pipeting bench after bench.

It looks like programming. And many of these companies actually, literally, to use their service, you write a computer program. So what makes us better is not just the fact that it's cheaper, but there's a huge disaster and, you know, real challenge in biology right now for reproducibility, that many biology experiments are just not reproducible. And it's crazy all the different variables that can come into play. But the beautiful thing about a cloud biology-like setup is that since it's programming on robots,

you have the best chance for reproducibility, that rerunning the experiment is rerunning the code. Making a small change in experiments, making a small change in the code and rerunning it. Sharing the experiment with some other colleagues is sharing the code. This actually now brings the ethos of programming into the biology realm, and that's part of what we say when we talk about it being better. One of the things that fascinates me, though, is that programming is programming, right? Code is code, and yes, it can be elegant, but then when it meets the world of biology or chemistry, people are complicated, right? Biology is complicated. How can code account for that?

Yeah, that's an interesting question. So I think there's often a misconception about what software can. do, you know, and I've often heard people talk about, you know, the intelligence of the machine and somehow it's going to magically replace, you know, human inside. And I'm here to tell you as a computer scientist that's not happening, at least not this week, right? So, you know, I see the software is something that brings great assistance to the researcher, right? And so what the software can do is look at billions of data points in an instant and provide some insights at what might be interesting to pursue versus what may not be. That's very different than having a human, sort of think about the creative process,

engage in sort of their knowledge of biology, and sort of interpret what the computer's providing to help make an ultimate decision, right? So for me, software is a tool. It's a very powerful tool. It's a way to accelerate the process. And I think also there's a part about software development that is this very iterative approach, right? And so when software developers go and build something, it's not like manufacturing a car where you produce it once and you hand it off, But rather, you make changes to that infrastructure weekly, daily, in some cases, right? And so that idea where you make a change, run an experiment, run a test, and do that very rapidly, right, in conjunction with doing things on the laboratory side and having these interfaces to these biological mechanisms, that allows you to iterate and learn very fast, right? And that's where the power of software kind of connects into the biological word.

Jeff, you straddle both the startup world and, you know, clearly your background in large pharma. How do those two, what Andrew describes, how ready is the large pharmaceutical industry to embrace that? Well, I think they're getting there and it's going to happen regardless of whether they embrace it. And I just want to add to what Andrew was talking about. The big data element of this is enormous, no pun intended. But right now, we have a huge amount or a huge number of data sets that for the most part are pretty siloed. when you can start to combine them and bring analytics to bear on them, the insights that you gain as compared to kind of the human trial and error approach of the past is huge.

I mean, one recent example is Watson Healthcare at IBM, I just bought a company called Treveen for $2.6 billion. It's a company that has cloud-based data sets for more than 8,500 life sciences, companies, and millions of lives that it has data on. When you combine those data sets with the other data sets that IBM has acquired recently and IBM Watson's analytical capability, you're going to learn things not just about drugs, but about humans that are beyond anything anybody had before. So going back to your original question, yes, there's always going to be a human element to this, of course, is always going to be interpretation and judgment and scientific discrimination and the rest. but the ability to have data from which to make these judgments well-structured, accessible, analyzed, interpreted is a game-changer

from where we were even, say, 10 years ago. At the end of the day, what we want is better health outcomes. So I wonder if you've seen this story before in another industry or maybe the other way to describe it is like what are we looking at and how are we going to benefit and what is that going to start to look? like. You know, actually, I would disagree that we don't just want better outcomes. We want better outcomes at a lower price. And it's a combination, which actually is the difference between sort of an advance and a really major step forward. You know, there's many different

industries in which compute has made huge advances. And we could think about things as the rise of machine learning in many different disciplines. And this is something which is obviously outside healthcare, but could make a huge impact. You know, it's really kind of shocking that computers now are getting routinely better than people. So, you know, years ago it was at chess, and now it's at go. Now it's an image recognition that computers routinely do better than individuals do. These advances are key milestones towards the role of and the capabilities of machine learning. But we even see this in everyday stuff.

If you think about companies like Lyft or Uber, what makes those interesting is not just that it's cheaper and not that it's easier, but that combination of that, you just have this cell phone network there and you can call a car and it can be sort of effortless. what creates his magical experience. And the analogous thing that could happen here is that the machine learning can allow computers to do things that humans just couldn't do before. But there's certain requirements. We need the data, and we need to be able to still have the right directions to point this to. And so the human part is certainly going to be there, but the opportunity is clear. Yeah, it's interesting. Yeah, data is something. That is my world, right? When without the data, we, you know, as computer scientists, just can't operate. And so there are tons of public data

sites available, the NIH Array Express in the European Union. So there is a lot of data that you can actually get your hands on. No agreement required, right? Millions of assays, just go ahead and download. We've also found that there's a lot of academic institutions that have data that they're happy to share with folks. But there also is just tons of data that's hiding in the silos of large farmers, right? And so we look at some of the farmers that we've interacted with, and they say, look, this is our goal. We're not handing this out. No offense, it's not you.

We're not given out to anybody. And we meet other companies like Sanofi that is doing more in terms of sharing data with folks. I believe that the way we're going to be able to, I think, make progress as a society and being able to process from this data and find new insights is through data sharing. And so whether that comes with some sort of business agreement or some way that we can aggregate those data and be able to give back to those who have contributed most in terms of the data that they have sought, But, you know, it's something that I think eventually is going to happen just because the benefits are going to outweigh, you know, the cost of just holding it tight. I mean, it's an open source kind of approach to it where large corporations can get involved and they can help facilitate the access to that data or the creation of that data and then they get the benefit in different ways. Yeah, or they don't even have to release the data. One can run AI on the data behind their firewall and use that to generate features or other things that could come out, such that I think there's a lot of creative.

ways to handle the IP challenges. I think another aspect of this that bears on pharma is that it's true that pharma has a lot of data, but the fact is the payers probably have more and better real-world data than the pharma companies do. And one of the reasons that the pharma companies will eventually, in my opinion, have to do data sharing, as Andrew suggests, is that the payers are going to aggregate their data, and they're going to start to have analysis and interpretation of what drugs work and how they work and who they work on that the pharma companies are going to have to contend with, and they won't have enough data by themselves to be able to do that.

So I think data sharing is inevitable, and I think, you know, you already see large payers aggregating databases, and as I said, in many cases, knowing more about the drugs and the drug companies. Right. One of the reasons Big Pharma is so big is because it's so darn expensive, right? And, you know, you see this again and again where a smaller biotech company has a promising molecule or something, and then it gets, you know, They can't push it across the finish line, so big pharma comes in and buys them and helps them get there.

Does it make the idea of big pharma less attractive? And does big farmer become smaller? Do we start to see a different industry? Well, I think you're on to something there. And I think if we were having this conversation five to ten years ago, the status quo you described was probably there. And if you think about it, 10, 15 years ago, pharma companies had drugs with 80% gross margins that were making tens of billions of dollars. growing at double digits a year. They didn't worry too much about costs.

You know, add another $100 million to a trial, so what if it's going to add billions of dollars to the drug sales? That's changed. The risk of drug development and drug discovery is increased. Therefore, the shareholders expect greater returns. Therefore, the pharma companies have to think about cost. And to VJ's point, it's not just better outcomes, but achieved it more efficiently. Meanwhile, another trend that's going on at the same time is the increased growth of small

virtual pharma companies that don't have all this legacy infrastructure that big pharma started with and themselves outsource and specialize themselves. So for example, I'm an investor in a specialty pharma company that has about $150 million of EBIT on. It's got 10 employees because it never built that infrastructure because like my movie analogy, it puts together each deal as it needs it with contract research, contract manufacturing, contract sales, and so forth. I think to your question, And pharma needs to get more and more like that, whether the big ones can get there, you know, it's going to be a major test of them, whether they can get smaller and leaner in order to grow. I think some of them are on the way to do that.

Others are still thinking in legacy terms. And also, I don't think it has to be mutually exclusive. I think there will be changes in pharma, but also there's just other models. We see actually really intriguing other types of models, models where foundations like the CF Foundation has played a huge role in pushing for new CF drugs. CF? Cystic fibrosis. Okay.

Yeah. And so that's an intriguing area because the cystic fibrosis foundation can actually do many things to help push something forward. They can help recruit patients and help with clinical trials as well as do funding for these trials. And so I think that's one model that is outside the traditional pharma model, but I think we'll see more of. Also, there'll be other cases where there'll be maybe more orphan issues where there is no drug at all. And so a clinical trial will be less expensive because the number of people you have to drive. it through would be a month smaller. There's not going to be the huge return that there would be for a blockbuster drug, but that the price performance will work out just fine for biotech.

An interesting little experiment that you can do to sort of test the degree of maturity with regard to outsourcing is to ask pharma people, what is the core competency of a pharmaceutical company? And what you discover is you get a lot of different answers. Some people say research, some people say sales, some people say commercialization, whatever. And the very variety of answers that you get suggests to me that they haven't quite figured out what their real core competency is. And in my opinion, the thing they're really good at that I find it hard to imagine being disrupted in the very near future is that translational stage from late stage assets to initial commercialization, figuring out what the market wants. And again, by analogy to the movie studios, that's what they do too.

Virtually everything else can be virtualized, no pun intended. And I think that will happen. And I think we'll see more and more different models, as VJ says. And the farmers that succeed are going to be the ones that outsource more and more of their cost centers and really focus on what they do best. Does efficiency in all these areas get us more toward personalized medicine, too? You're talking about orphan drugs. But, I mean, there's always an orphan for a drug, probably. Yeah, you know, there is some really tantalizing possibilities here that 80% of all cancers could be cured with existing drugs. if we caught them early enough, and we knew to give them early enough.

And so what we're also seeing is huge advances on the diagnostic side from advances in genomics or proteomics and metabolomics. And that clinical side, actually, I think will interface very nicely with what we're talking about. And the diagnostic side will interface nicely with the therapeutic side. And that could be a very different world. I mean, the challenge there is getting diagnostics to be considerably more accurate than they were before.

But that's an interesting challenge. And also, that itself is an interesting machine learning and computer. science challenge, to take these new data sets, to combine what you'd get from proteomics and genomics and other, take everything you can get and make the most accurate predictions and then give the best drugs we have now. That itself would be a different role that doesn't require anything on the pure pharmacide. Yeah, and I think we need to really over time, also build enough data to be able to pull those things off. I mean, like for me, ultimate personalized medicine, as I walk into my pharmacy, I say I'm not feeling well,

biological samples are collected, data is generated, and I'm formulated a solution right on the spot, right? That's ultimate personalized medicine. And so we can't do that today because the ability to generate the enough quantity of data to be able to make a prediction to give someone a drug on the spot, just the costs are too high, right? And so we as data scientists are sort of looking at larger populations, and we're starting to go down the path of being able to do a subpopulation on disease and individuals. And so eventually, I think, will have enough data that those sorts of things become very easy to do. But for today, right now, I think we're just, we're more headed down that path rather than that's something we can pull off

at the moment. I mean, as a, as a healthcare consumer, that sounds great. I walk into a pharmacy, I walk out with, you know, what what I need for what ails me, the sort of hurdles liability-wise and otherwise, is that a scenario that we get to easily or easily is the wrong word ever? Oh, well, we definitely get there and not easily and not soon. But everything, is headed in that direction. And if nothing else, the customers and patients will demand it. I mean, we have a generation of people today that are used to being able to get services and goods on demand very quickly, very inexpensively, without a lot of intermediaries. And there's no reason they're going to accept the kind of health care system that their grandparents lived with,

which is functionally what we have today. So I think we will get there. There will be a lot of disintermediation. There will be a lot of companies that will fail as a result of this. It probably will have a salutary effect on pricing. And, you know, there's an absolute need for it. And I think history suggests when there's a customer demand and a patient need, it ultimately gets filled. I think health care has historically been, and I hate to say it's different from other industries, but in some respects it is. It's an incredibly complex ecosystem with, you know, thousands of different interests, sometimes

in conflict, whether it be hospitals, doctors, payers, patient groups. groups, pharma companies, the government. It's also been a system that historically has had intermediation between the customer and the ultimate provision of care, drugs being just one example of that. And it's been an industry where there's been, understandably and rightly, a huge focus on quality and control. And as a result of those things, it's been a little bit slow to change. And just to give you one example that I think people in the tech industry might find remarkable, it was only relatively recently that the FDA allowed clinical data to be entered directly into computers as opposed to put on paper because there was a

reluctance to trust the quality and integrity of that data, which I think we would all agree is kind of silly because it's much more reliable than paper. But just the fact that that only happened recently suggests that when you're in an industry with a regulator that is quite rightly and understandably concerned with quality, it can be slow to change. But I think that change is going to accelerate. I think it's not going to be long before we see things. along the lines Andrew's talking about.

Yeah, and I think, you know, it's really exciting about the future of having data sciences get involved in some of these things, and especially we're talking earlier about preclinical studies and animal models, which, you know, as it turns out, a mouse is not terribly predictive of what might actually happen in a human being. We've got cancer in mice a million times. Exactly, right? And so when we look at some of the data models and what we can do in terms of the predictive power of those models, you know, I see a future where we're getting rid of the animal

because what we can simulate in the computer is going to do a better job of predicting what's to happen in that human being, right? So what does that mean? We're talking about, you know, the FDA, right? We're going to convince the FDA, we're just going to put it all on the computer and then we're going to drop it in a person, right? Like, that's the types of changes that I see coming in the decades to come. And the good news is we're building that capability and that that is part of our future. But to your point, it's going to take a while to get there, right? There's a lot of hard evidence that we're going to have to produce. We're going to make sure that, you know, these things are safe and reliable. But it is where we're going, right?

There's no doubt. Also, for many, the reputation of Silicon Valley, is something like move fast and break things. Which when it's your internal organs involved, yeah. Yeah, yeah, yeah. And so, and I think it's important to emphasize that even though there's very much a lot of the Silicon Valley culture of applying computation and innovation involved here, it's understood that this is a different type of space and that there's going to be new types of challenges. I can't wait to get into my driverless car and go to the pharmacy and stand in front of the machine and it spits out a pill just for me. And maybe you won't have to go to the pharmacy.

Yeah, I think it's a blood-tested home and something arrives later that day. Even better. Jeff, I don't mean to paint you with the big farmer brush, but I'm going to anyway. You and Andrew are sitting here side by side. Andrew is definitely a product of computer science and here in Silicon Valley. How is it that you guys are getting along, not in this room, but how is the culture of where you come from, Andrew, meeting the culture of, you know, Jeff, that you used to inhabit and still do? Well, in the early days and maybe even to some extent today of interactions between the Valley and pharma, I'd be in rooms where those two people were meeting and one speaking Greek and one speaking Latin. There was just a fundamental cultural divide. They didn't understand each other's business and industry. Andrew can speak to whether or not he experiences that today. But I think it's got to change on both sides. You know, I used to, when I first started doing investing in this space 10 years ago, I would see proposals.

from Silicon Valley that made no sense from a health care perspective and a total failure on the health care side to understand the value of computational benefits. So I think that's changing. I think it's got to change like it has in every other industry. You know, the history of industry in general is that the disruptors usually come from the outside. And one of the challenges pharma has had is its own success. It hasn't had an existential crisis, the way IBM did in the Gersner era, for example, which is one of the few examples you can think of where a big industry really changed. Most of the time, it's Amazon taking over Barnes & Noble. So I imagine, especially going back to my comments about virtual pharma companies, I imagine a lot of this innovation

will come outside of big pharma, but the big farmers that succeed will be the ones that get it. Andrew, yeah. How is it when you sit down in the halls of big pharma? I'm going to put it out there right now. I'm happy to admit this. I'm a weird dude. So, you know, when software engineers and mathematicians think about problems, we think about things in just way different ways. And so one of the things that I think is just so interesting is that self-driving cars did not come out of a GM or a Ford or an automobile company. It came out of a software company, came out of Google, right? And so that way of thinking, that culture that, you know, sort of stepping back from the problem and just twisting it in a very different way, I think is quite compelling, right?

There's just sort of this different view on problems that I think, you know, can offer insights and offer new abilities to produce things that maybe others haven't considered. And it's almost some of these things feel like a matter of time, but there will be hiccups. You know, before Amazon, there was Webvan. And after Webvan, you might think, well, you know, this Internet thing is not going to work out. Like, you would never be useful for home delivery or anything like that. but lo and behold, comes a company that takes these advances at the right place and the right time and does something great. There's almost like a deterministic nature of Moore's law, that the cost of compute and cost of genomics and these things decreasing exponentially, that actually is something that's really hard to avoid.

And so in my mind, it's not really a question of if, but when this happens. And that I expect actually the next 10 years could be really fantastic towards those ends. Vij, Jeff, Andrew, thank you guys so much. Thank you. You're welcome.