a16z Podcast - a16z Podcast: Taking the Pulse on Bio

Episode Date: December 14, 2017

This conversation between the members of a16z's bio team -- including general partners Jorge Conde and Vijay Pande; Malinka Walaliyadde; and Jeffrey Low (the interviewer) -- takes a quick pulse on where we are with when bio becomes more like engineering. Especially given the announcement of our second bio fund, this episode of the a16z Podcast covers everything from the broader trends at play to some specific areas of interest... as well as what types of entrepreneurs may bring us forward into the new Century of Biology.

Hi and welcome to the A16Z podcast. I'm Hannah, and we're here today for the first time ever with the bio team all together to take a pulse on where we are on the intersection of bio and engineering and what it makes possible. The conversation includes general partners Vijay Pande and Jorge Condi and Malinka Wallalliade, interviewed by Jeffrey Lowe, and covers everything from what the shift away from empirical science towards engineering means for bio to what we're looking for in entrepreneurs now that we've announced our second bio fund. Today, we're going to talk about what we've seen at the intersection of biology and computer science and how engineering and biology is changing how we think about bio investments. Let's start with computational biomedicine. Vij, how do we think about this at the start of the first bio fund? Actually, this was very much inspiration for the first fund in the first place, that we saw the existence of companies that are really tech companies at their heart, but that can be built in the biology

and healthcare space, and that machine learning was a key means towards that end. And what have we seen over these past two years? One way to divide medicine up, a traditional way, is between diagnostics and therapeutics. In the diagnostic space, here there's a very natural trend that you take some new data source, whether that would be genomics or wearables or other new technologies, and then you marry that with artificial intelligence or machine learning. Some means to go through all this data and to gain insight faster and higher accuracy with continued learning in a way that we really couldn't do before.

And then finally, towards some ends that is actionable. So a great example of this is something like Freenome where you take genomics and the data from what the DNA in your blood tells you about your immune system. But, you know, we don't understand the immune system. So we use AI to be able to sort of tell us what this means with high accuracy. And then it's very actionable that you would get the appropriate cancer procedure done, especially if you could catch cancer early. This is where AI could really be a part of the key missing link towards the killer cancer. Two years ago, there weren't that many examples of prominent journals publishing AI and health care, where it was shown to work in healthcare contexts.

And now it feels like every month there's a new nature paper or something coming out where it's been shown to be demonstrated to be much more accurate than a human physician. The application of AI or machine learning actually doesn't allow you to just do something better or faster or cheaper. It actually allows you to do something that previously was impossible.

One of the biggest challenges in developing therapeutics has been understanding where to target, specifically a disease, what mechanism, what target to go after, and how to develop chemical assets against that. And I think one of the things that machine learning can help us with is to unravel the inherent complexity of biology in a way that we're not depending on our human understanding, our human minds understanding of the complexity of the disease in order to determine how best to intervene. So diagnostics, therapeutics, these are two areas that we've seen AI applied.

Is that where the future is or is it in a third space? It's really just one subset of what we think is. the broader theme here, which is the shift away from biology being primarily an empirical or an experimental science to becoming more of an engineered discipline. And so machine learning, artificial intelligence is one example of the application of an engineering approach to biology, but there are many others. And what do you mean by an engineering approach? We lack a sort of fundamental foundational understanding of what drives most biology, and specifically what drives disease biology. You have to essentially say, I have a hypothesis that this particular target

may be the optimal point of intervention for, you know, for intervening in a disease process. And that requires experimentation. And as we know with experimentation, sometimes it's successful. Oftentimes it fails. So when you take more of an engineering-based approach, it's the discipline of saying, well, let's take the science risk out of it. And there are many ways that you can do that. One way, as we just covered, is to use machine learning or artificial intelligence so that the algorithm itself can learn the underlying biology in a way that we can't as we just covered. But another way is to focus on areas where the biology is relatively well understood and now find

ways to improve our ability to intervene. So one example on the therapeutic side might be in areas where the cause of the disease is very well understood. So in sickle cell anemia, it's long been understood that the cause of the disease is a mutation in hemoglobin that causes red blood cells to fold or sickle. And so if you wanted to develop a therapy for sickle cell anemia, what you would need to do is to find a way to either replace or repair hemoglobin so that your red blood cells don't fold. And there are many engineering-focused approaches to doing that, namely CRISPR or gene therapy,

that are focused on a very specific way to solve what is already known as the biological problem. To give you a contrast to that, take a disease like Alzheimer's, where we actually don't really understand what is driving the disease. And if you don't understand the biological basis of the disease, then you have to hypothesize on what you think might be the best point of intervention. And that's very challenging. That's a very science-based approach. You experiment, you test. And as we've seen very often, unfortunately, many of the therapies that have been developed for Alzheimer's fail. And often they fail very late in the process in a phase three trial, which of course is very expensive and very time-consuming. And so that's how we sort of differentiate is how can we think about focusing on opportunities where there's

not a lot of science risk, novel biology risk, and focus instead on opportunities where there is perhaps engineering and scale risk. And that's an area that we think we're well positioned to help support entrepreneurs in that space. For decades, anything in this biospace, biotech space would be really dominated by how to mitigate science risk. And that the new opportunity now is through new technology, either machine learning computer or otherwise, that there are more and more areas where that science risk is no longer the issue. issue. And that's the opportunity I think that we're looking to go after. It's remarkable to have seen how broadly applied it is. So we've seen this in dermatology,

analysis of the retina. We've seen this in applications of actually using facial features to identify genetic diseases. We've seen this in analyzing pulse from an Apple Watch to diagnose heart conditions. So this is a remarkable C-change in the way we think about how we can collect data and make sense of it using this intersection of sensors and artificial intelligence to really make sense of what's driving disease. The other big changes is that when you have these new data sets, AI allows you to create whole new areas, much like you'd create new apps. So, like, you know, in the old days before the smartphone, you had your GPS box and your

camera and your phone, and you'd have to have a whole not a piece of hardware. Similarly, in diagnostics, if you have your PSA test, you're not going to do a BRCA test or something like that. Genomics could be used for so many different things. If you have data for wearables, it can be used for so many different things. And often, it's not sort of whole new. science. It's just repeating the exact same engineering process you had to get the first test for freedom to do one type of cancer and do a second type of cancer from genomics is really essentially

running the exact same. And you can contrast, you know, a company is able to develop a completely new test from its data compared to old diagnostic companies that, you know, have to develop new things. You know, science is something that you can't schedule creativity or the ability to come through a breakthrough. So if you have understood the science of colorectal cancer, what you've learned about a stool test probably is not going to be very useful for a breast cancer test. It would be a completely different biology. But if you're using genomics and you can be able to learn from training from samples, it's basically just getting new data and new samples and repeating the exact same process. I mean, this speaks to sort of the different ways in which

this has shifted from science engineering and the ways that, you know, engineering and AI makes it more accurate and more repeatable. There's also been non-artificial intelligence means of engineering. Where has engineering and biology come in that has not involved artificial intelligence? One of the key areas where we see biology itself as a tool for engineering has, of course, been in the gene therapy space, in the CRISPR space, the technology that allows us to do very precise gene editing or genome-wide editing. And we've also seen at the cell level, at the cell engineering level, you can now use these very precise tools to either replace DNA or edit DNA, or in some cases even write DNA, and that's very much an engineering

driven effect. Take, for example, next generation sequencing for DNA. Today, because of improvements that were born of the application of engineered disciplines, and three very specific ones, we can now sequence a human genome in what used to take 13 years in a matter of hours, and what used to cost $3 billion for less than 1,000. And really what drove that wasn't some fundamental discovery on the biology or the science, what drove that was aluminum, was able to apply three very specific and distinct engineering disciplines and converge them. One was the use of microfluidics, just a fancy word for saying, to use piping to move around tiny amounts of liquids and chemicals so you can run experiments in a very, very dense environment. The second one is the use of optics,

so you can in fact detect chemical reactions, again happening in very dense environments. And the third one is the use of compute to enable you to actually string together all of this data to actually recreate what would be a representative genome. And it was really those three engineering-based disciplines that gave us the ability to do genomics as it exists today. It takes something from being impossible to now possible to eventually becoming routine and then from there indispensable. We see the same thing happen in applied engineering approaches to how we design monoclonal antibodies. We've seen this now with MRNA and the ability to produce that on demand. And we've seen this again and again. One of the very interesting things about this particular space is these companies will look

very tech-like in terms of their ability to create new markets and scale and dominate those markets. And those are obviously very attractive and interesting opportunities for us. One really interesting aspect on the engineering biology side is how there's this entirely new wave of cell therapies that's coming out, which is an entirely new modality of drugs. I mean, we started with small molecules, then proteins in the 70s or 80s with genetic, and now there's this entirely new modality coming out, which of course means that there's a lot more demand for types of delivery methods and in terms of tools for this modality.

Yeah, we had small molecules, then we have large molecules. Engineered cells represent essentially living drugs, which is a remarkable advancement. And I think one of the things that we're seeing is that now when we can start to design cells, we can actually start to design cells that have logic, that will know where to go in the body, what to do when they encountered disease, and how to essentially terminate themselves when the disease state has been resolved. So we're on the cusp of not only going from having living drugs, but having intelligent drugs. And that is an incredible step forward. Well, let's move on to digital health.

What's the state of the field of digital health now and where's digital health going? One of the huge opportunities there is that one can actually, first off, design therapy, that don't have any toxicity. When we think about a therapeutic, we can think about something like an antibiotic. And an antibiotic is kind of a magical thing, right? Take this pill, and like a week later, you're done and you're fine. But there's no equivalent for that in so many areas. There's no pill that you take for a weekend.

You have no more PTSD or you have no more anxiety or you have no depression or you have no more type 2 diabetes. So what these things have in common, first off, is that the biology is very complicated. It's not something where there's some invading pathogen that you need. need to get rid of. But we actually do have therapies for them. They're just not pills. These therapies are behavioral therapies and often have pretty good efficacy, but they're very expensive and hard to scale. A great example of this is the CDC's diabetes prevention program that existed and was validated. The science was de-risk by the CDC and others. And so that was very appealing.

And companies like Omata now can take that engineer it, scale it. And the intriguing thing is that they can do the equivalent of clinical trials, except in computer. computer land, the clinical trials, AB testing. And they can do this AB testing, you know, if they wanted to once a week at scale. They can constantly iterate to make their therapeutic have higher efficacy without having any issue with toxicity. The challenge has been very much to find areas where you could make that similar impact in areas other than type 2 diabetes. We've seen network effects as an area that has changed a lot of the tech ecosystem. How have those played out in the healthcare space? We've talked about network effects at length before, but the quick gist is a network

effect is any company that's able to build a network where the more people that join the network, the more powerful and defensible that company is. There's a lot we've learned in the pure tech, consumer tech side of the world in terms of network effects that we've seen in the biospace. So an interesting example is this company called Patient Ping, which there's something called care coordination. You know, hospitals are increasingly taking financial risk on their patients. They are financially on the hook for excessive care. Care coordinates at this hospital actually take risk on patients who leave the hospital and go to external facilities, but they have zero idea where these patients actually go. But somehow they're

financially liable for these people, which is insane if you think about it. And so what patient ping does is they actually connect all these providers together and care coordinates in the hospital and other places get a ping anytime one of their patients goes and checks in somewhere else. It's a very simple product. But what's really powerful there is they form a really, really powerful network effect. And the more providers join the network, the more valuable the network is because the more granularly they can track their patients. And so this is very much a concept from the traditional tech and consumer world that's been brought into healthcare. It's difficult to develop, but when they do happen, they're very powerful. That's it. Now, this is just one type

of network effect. This is a people network effect. There's others, of course. The real heart of network effects is to try to create two things. One is a barrier to entry, but also there's this powerful effect that, you know, as you get more customers, this barrier to entry grows. And one way to do this is through our so-called data network effect. And we've seen this with companies like Freenome or Cardiogram in these diagnostic spaces where as each of them gets more data, they get more accurate. The tests go from like 85 to 90 and 95, 97 and hopefully even higher percent accuracy. And then as they get higher accuracy tests, they get more customers because they have a better product, which of course gets some more data and it just spins up from

there. What's intriguing is that this is all about barrier to entry. And in traditional biotech barrier entry is largely due to patents. And the patent window is shrinking. It's such a challenge. But, you know, data network effects never go off patent. They just get stronger and stronger and help companies grow even after, even decades after. One of the things that I've seen since joining is we're looking more and more into therapeutics. We're looking for companies where the science risk has been removed or greatly de-risk. And so that often sort of removes a lot of traditional type of therapeutics companies, companies where there is a huge amount of science risk. But I think there will be examples of therapeutics where are driven by a much more engineering approach, whether we're talking about engineering cells, using AI for new ways to find new small molecules for biologics.

I think there's a lot of potential there. What's fascinating about what we would call traditional therapeutics is that it's historically been a very sort of bespoke effort. So you're focused on a disease area. You try to identify a target where you want to intervene. And then you make a very specific thing, a specific molecule that will hit. that target, and ideally only that target. And so by extension, the next time you want to make a second drug or another drug, in many ways, there's not a whole lot you can take from your experience from having discovered and developed the first drug, because again, it's a very bespoke thing.

So I think one of the things that we think about from an engineering-based approach is where are there examples where you can actually transfer knowledge and experience from the first to the second to the third and so on. And the application of a very example. AI and drug discovery, I think is one example. And I think there are other examples that we look at that are biologically engineering-based approaches, whether it's, you know, CRISPR or MRNA or any other number of things where the tool is a biological tool where really you're using a modality and you're just swapping out, you know, the code that you're actually inserting into the drug in the case of CRISPR or into the case of MRNA. And so those are the kinds of things that I think

are potentially more interesting to us because it takes drug development from being a very bespoke thing to being something that's more generalizable once you understand the underlying disease biology. Are you interested in modalities where the tool like CRISPR is the therapeutic itself, or are there other ways where CRISPR or RNA could be involved in a therapeutics company? Well, I think CRISPR is a great example that shows that innovation in our space is accelerating. CRISPR is a concept barely registered just a few years ago. And now it's an indispensable tool for drug discovery, actually for understanding biology.

as we've seen the cycles of iteration in biology accelerate, as we move biology more and more into an engineered-based discipline, I think we're going to see more powerful modalities like CRISPR. And in fact, CRISPR already has emerged into various different flavors. And I wouldn't be surprised if we see sort of the next new thing emerge. Just think about what's emerged over our last two years in biology. It's kind of breathtaking, you know, compared to, let's say, the previous 10 years. And I think we're going to continue to see those opportunities emerge. Okay.

As we go from hype to creating real companies, these companies all will have to go to market. So what's been working? What channels have been working? And what have these companies been doing? There's a couple different stages. And you have to do with different types of people that you have to get to. So first off, often you're dealing with CMO and you have to convince him or her that this actually works, demonstrates some degree of efficacy of a clinical trial, some other study or something to point to.

But then actually beyond that, then you have to get to the point where you get reimbursed and demonstrate there's real value. And that's the next step. That's when you can start to get commercial traction. So we start with evidence, then go to reimbursement. And yet, you know, who are they getting this money from? A very common pathway when we first started was going to self-insured employers. So most employers over 500 employees will self-insure. What that means is they hold the financial risk.

The employer actually pays for their employees' health care. And then they work with a third-party administrator or an administrative services organization, like an Aetna or Cigna or whatever to actually do the claims processing on the back end. When a company gets large enough, they can self-insure and hold the financial risk. So a lot of startups started with self-insured employers, and that was actually a pretty popular approach early on.

And here's why they started with self-insured employers. It's because as a healthcare company, when you sell it to pairs, when you share a healthcare service to payers, there's two different value props that you can sell against. One is you actually save them money on healthcare costs. and the other is it is an engaging product that can be used by their members or their employees, which makes those members or employees like the employer or health insurance plan more.

Now, health insurance plans care a lot more about saving money than offering an engaging product. And so for an early-sage startup, it's a lot harder for them to do the saving money piece because that takes a long study to prove, and they do it eventually, but it's hard to do that initially. But they can show that it is an engaging product early. And therefore, employers who cared about both of those pieces a lot, they can go and sell to employers on the engaging product piece. And that's why healthcare self-insured employers were often considered early adopters. Also, you often had very forward-thinking benefits people at these self-insured employers.

For example, Sean Levitt of Comcast, who's been, you know, long considered a thought leader in the benefit space who will work with startups early on and, you know, distribute those benefits to the employees. Now, unfortunately, that approach got a little too popular and those channels got a bit too full and benefits leaders were getting pitched by these vendors all the time. So at this point, there's really a few large digital health and startups in the space that have been able to deploy at those employers. And the next step now is for them to actually start working directly with plans and distributing to those plans,

because you get obviously much, much wider distribution. Fortunately, many of the administrators at plans are now also much more sophisticated about this space. It's just a matter of time, and they're working directly with startups do. And this is something we've seen through Amada as well. So once you've gotten through the early adopters on the self-insured employer side, you then to get to the mass market need to go to the regular insurance plans. Yeah.

Yes. It reminds me of the saying that, you know, the federal versus state, that the states are the laboratories of democracy before it actually gets up to the federal level. And I think that's probably true in this case. And that's actually a good point because that once you go through plans, actually the next step oftentimes is working with CMS and going through Medicare or Medicaid. Now, people can change those steps up, but that's a progression that we've seen. Yeah. So it sounds like, you know, a lot of times they go from the insurance plans to the government, to CMS, to Medicare and Medicaid.

We've also seen a lot of developments in other part of the government, namely the FDA. You know, there have been a lot of new regulations out, the first FDA-approved digital therapeutic, the new programs for pre-certification. Can you talk about how these might have affected some of the companies and investments that we might make in space? Well, I think from a regulatory standpoint, we're going to see across the entire spectrum incredible advances in terms of what's happening at the FDA level. We'll see more generics approved this year than in any year in history previously. We saw this year the first cell engineered cell therapy approved in Carty. Historically, you know, FDA risk or regulatory risk was really used as this barrier and this essentially fear factor. But I think as the therapies become more powerful, as we,

target things where we understand fundamentally what's driving the disease biology and therefore understand fundamentally how best to intervene, I think what becomes pretty clear that regulatory risk in many ways was just a euphemism for scientific risk, for experimental risk. Gene therapy, an area that historically has been considered to be very risky, has been shown to be so effective in this form of treating an inheritable form of blindness that that passed the FDA panel recommendation unanimously, a 13 to 0 vote. Carty, similarly, also passed the FDA panel at a 13 to zero vote. I mean, those are two things that, two modalities, two therapeutic modalities,

two treatment modalities that were considered to be very, very risky. But they were shown to be so effective that that sort of risk goes away. And I think when we see a lot of these therapies emerging, the benefit, the effect is so powerful that it's very clear that the therapies work and therefore the regulatory risk is low. So traditional therapeutics companies have gone to market. through a regulated process of clinical trials, and they've kind of looked the same way for quite a long time. As we see this new type of biocompany, will those look any different? There are emerging technologies that are going to change the way we think about testing and ensuring that our drugs are safe and effective.

So just to take a step back in history, or at least to where we are today even, the way we ensure a drug is safe and effective is first we determine whether or not we think a drug might work. And the way we do that is by testing it in animal models, and cells in a petri dish. And once we've convinced herself that a drug might be safe and might be effective in those models, we then go to human clinical trials where we, of course, need human volunteers

to confirm that a drug is in fact safe and effective. One of the big challenges that we've seen historically is that that paradigm is not particularly effective. We still have massive failure rates in clinical trials. We still have massive failure rates in phase three clinical trials, at which point hundreds of millions of dollars and years and years of years of investors, have already been made. And a part of that is because there's so much science risk in some

of the therapies that have been developed historically. But a part of it is also that the preclinical models, using animal models, using cells and feature dishes, is not a particularly effective way to predict if something's going to work in a human being. And one of the exciting things that we're seeing emerging at this intersection of biology, of engineering, is new approaches to essentially obviate the need to use animal models as a replicate, as an avatar. for what might work in a human, and we're seeing the development of organs on chips or eventually humans on chips that will allow us to really test something in a more human-like engineered system that we hope and believe can be much more predictive of what, in fact, will happen

in a human being. So I think that part of the paradigm we're starting to see shift. The other part of the paradigm we're starting to see shift is that historically clinical trials have been very large because you needed to recruit a lot of patients to determine whether or not a drug would work. With more targeted drugs, we need less patients because the signal to noise is likely going to be higher. The second thing that's been a big challenge in clinical trials is that they take a very long time to find the patients, to recruit them, to get the center set up. And this is an area where we get a major assist from technology. A great data point that I heard recently was a partnership that Facebook made with the Michael J. Fox Foundation to find ways

to pilot the use of social networks to recruit patients for Parkinson's disease. trial. Using social media, they were able to reduce the cost and the time associated with recruiting patients by like 96%. That's transformative. And so we start to see compression how we ensure that drugs are safe and effective and they start to become more predictive. We're going to see failure rates go down and we're going to see our ability to innovate in therapies accelerate dramatically. So traditionally we've seen tech investing go for companies where the technical risk was quite low, but the market risk quite high. They're investing in growth and they see profits over time, biotech investors looking at places where a science risk is high, but the

market is known if you make that drug, it's going to sell. If we see this new type of company, how do tech and biotech investors come together and work together or work separately on these kind of companies? You know, where we live is really at the intersection between the two. And so obviously there'll be the need for connecting on the deep technical domain of whether it be computer science or machine learning as well as on the biology side. Collaborations are cases where there may be some major biological advance that shifts into the engineering area. Now, these become tech companies or that biological advance comes hand in hand with a machine learning advance and they synergistically be able to push things forward.

What we're really seeing emerging are new founders that also live in both of these worlds, founders that have deep experience in the biology and deep experience in computer science. And actually, for these types of companies, I think that's going to be critically important. You know, you can have companies where each of this expertise are in two people, But, you know, that only works if these two people are basically can read each other's minds and can be telepathic. Having in one person really changes the game, and I think that's going to be one of the major trends going forward. Okay. Let's do a lightning round of new technologies on the frontier.

What's the coolest new technology that you think will change biology in the future? You know, we spoke so much about engineering biology. I think what we're going to start to see is the literal engineering of biological circuits inside cells, the ability to design circuits, much like we design electronic circuits. And the reason why this is important is that there's only so much you can do by hand. And by having essentially all the lessons from the electronic design automation, EDA, tool flow on the electronic circuit side, applied to biology, could really unleash the same type of effect that we've seen in electronics over the last 50 years, but now in biology. Yeah, and I think just to play off that, I think one of the most fascinating things that we've seen here is how biology is no longer an industry in and of itself. It's in something that's going to touch every single industry. And we're already seeing applications in energy, in textiles, in food, of course, in health, and in data storage.

And so I think computation itself is at some point going to be affected by biology. And I think it's a wonderful time to be an entrepreneur and an innovator in this space. Melenka. What's the craziest thing you can imagine? EHRs that actually talk to each other. Epic and being able to be in a medical record between epic and zernet. Dream big. I also think we're going to see a lot of interesting things.

in the non-therapeutic CRISPR space. So in a therapeutic CRISPR side, there are a bunch of public companies that are going directly after using CRISPR to cure disease. I think it's just the beginning for non-therapeutic CRISPR. That is, using CRISPR for areas like diagnostics to find out infections, using CRISPR to discover new drugs as a platform for discovery rather than the direct treatment of disease itself. And I think we're just starting there.

Thanks so much for joining the A16Z podcast. This is the A16Z bioteam, signing off. Thank you.