a16z Podcast - a16z Podcast: When Bio Meets Computer Science

Episode Date: July 26, 2015

Biology startups have been around for a long time. But the world has changed since that first wave of bio startups, and especially more recently, due to the "peace dividends of the smartphone wars". So what happens when you combine those cheap sensors and compute power -- and apply it to bio? Cheaper costs and Moore's Law-like effects may mean lower barriers to entry, the ability to experiment more often and more easily, and other AWS-like effects for a new wave of bio company founders. But is all this just about cost … or about something more? And how do we know if this time is really different when it comes to bio startups? On this episode of the a16z Podcast, a16z partners Marc Andreessen, Chris Dixon, and Vijay Pande discuss how everything changes when software eats biology -- that is, when computer science is at the heart of bio companies. Pande also shares three trends we think are particularly interesting: "digital therapeutics", "cloud bio", and "computational biomedicine".

Hi everyone, welcome to the A6 and Z podcast. Today's episode features A6 and Z partners, Mark Andreessen, Chris Dixon, and Vijay, who spent the last year here as a professor in residence. The topic we're focusing on today's bio, and more particularly, the intersection of biology with computer science. We're starting to see an inflection of many different things coming together, and this is a confluence of many effects on the biology and genomic side as well as on the computer science side as well. Clearly, the world has changed. So what's different? Why now? Is more bio-innovation possible for more entrepreneurs than ever before because of cost, much like what AWS did for web startups, or is it about something more?

It's not just about cost or capex. It's about doing things that you couldn't do before. We're seeing the ability to do types of experiments that even with a huge pile of cash wouldn't necessarily be all that easy to do. And so having something which, in a sense, used to be impossible, now become cheap, inspires people to do really new and exciting things. And finally, how do we know? As it many innovations that come around before but take off only at certain times, how do we know that this time is different?

You say it's different this time. I'm skeptical. Yeah, no, that's very natural, I think, because things are always different this time until they're different. Okay, so that's the introduction for today's A6 and Z podcast. Let's hear from Mark, Chris, and VJ. Let's start by thinking about history and then think about what's going to happen in the future. So historically, you know, for the last 30 or 40 years, there have been kind of two worlds of venture capital and two worlds of startups, high-tech startups. There's been the computer science or IT world, which is well known. And then there's been the life sciences world. And the life sciences world has been characterized by, you know, terms like biotech, new drug companies, and then also new medical device companies.

And so, you know, technology stents and all kinds of new, you know, all kinds of new implantable medical devices have come out of that. Historically, these have been two very different worlds, very different founder profiles, very different. In fact, there's been very little intermingling between the two worlds. historically. And that's sort of the last 20 or 30 years. Do we think that's how things will continue, or do we think that things are changing? And if they're changing, why are they changing? Yeah, we're seeing something that is really a pretty significant change because these two worlds are really not separate anymore. You know, you think about actually students at Stanford, some huge fraction of them, I think, like 75% or more take some sort of computer science class,

some programming computer science class. So we have a new breed of biologists and chemists and doctors who not just are familiar with computers, but actually they know how to program, they know the details. And so these two camps are not necessarily separate anymore. We're starting to see something where it is a combination of the two, which is actually an interesting challenge because it sort of requires expertise in both of these areas,

which themselves are each quite deep. Right. And how recent do we think this has started to change? Yeah, I think it's like anything, it's some of these things where it grows slowly and then sort of hits an inflection point. And I think just over the last year or so, we're starting to see an inflection of many different things coming together.

And, you know, this is a confluence of many effects on the, on the biology and genomic side as well as on the computer science side as well. Take us through your view of, like, what is a modern bio startup or a modern converged bio and CS startup? Like, what are the characteristics of that new kind of startup? Yeah, there's actually several things that we're seeing. And I think actually these startups have seen the success on the software side and are starting to borrow a lot of those ideas. So one key thing is that they can move very quickly and move quickly with a small amount of initial capital. And this is something that's really quite new. It used to be that when you start a bio startup, you'd have to put a lot of money into building up a lab

and building up a large team to run that lab. And that's something that's really radically different. The second thing is there's just a lot of information. And so, A, that's an opportunity, but it's also a challenge. How do you handle all that information? So in a lot of ways, it's similar to what happened to semiconductors in, I think, around 1980 or so when you had the advent of fabulous semiconductors, which meant that instead of a semiconductor company having to fabricate their own silicon, they could outsource it to places like Taiwan, and then just a very

small group of engineers with little capital could create an innovative new semiconductor, and that created an explosion of new startups. And then we saw something similar with internet startups in the 90s you had to spend, raise tens of millions of dollars, set up your own data center, you know, build your own kind of back-end systems, and then you had things like AWS, and, you know, famously, you know, you have Instagram with a couple, you know, 15 people and billions of users and WhatsApp and et cetera. It seems like we're seeing some of that happening now in biology. Yeah, I think it's exactly right. And it's interesting. There's two ways this is having an impact.

One is the obvious that people are looking on the biology side and seeing that one can do things this way. And so, in a sense, you know, these young biologists are all. on Instagram, they've seen how this thing has worked and are very much inspired by it. But the second one is that there's this great appeal to move very quickly. And so just like you can move quickly with AWS, nobody really wants to build up a lab and do all that. You want to try your idea very quickly and see how it goes. But then the final thing I think, which is to me the most important one, is that it's not just about cost or capex. It's about doing things that you couldn't do before.

We're seeing the ability to do types of experiments that even with a huge pile of cash wood, necessarily be all that easy to do. And so having something which in a sense used to be impossible now become cheap inspires people to do really new and exciting things. So let's, yeah, let's talk about some categories. Let's talk about some examples. Say you're at a dinner party or perhaps on a podcast. And you want to go through, let's maybe go through three or four examples of categories where we think that this is happening. Yeah, so that's a great question. And I think there's a couple different categories that we're seeing. So one category that is, I think, a natural one because of all the devices and social networks that have been built is an emerging area of

digital therapeutics. And, you know, what's intriguing here is that there's certain cases where it's very natural that the current medical approach is really the natural approach. You know, if you are in a car accident and you're bleeding, you know, there's certain surgery that you do or if you have bacterial infection, you want to take an antibiotic. But similarly, there's actually some things that I think it's natural to question whether a drug is really the natural approach, areas such as in depression or sleeping issues or smoking cessation. And these more lifestyle issues, which are starting to become real dominant issues in our culture today,

it's not clear that a drug really is the best approach. And what's intriguing is that maybe now with the emergence of digital therapeutics, with the fact that we all have phones on us, all the time that can sort of detect and understand what's going on with us. Combine with social, taking that data and connecting us to other people allows for opportunities in many interesting areas in these places where maybe drugs would not work as well. So let's dig into the term a little bit. So digital therapeutic, meaning that it is a solution to a medical problem.

It's some level of cure or treatment for a problem. Digital meaning it's not a drug. It's not an implantable device. It's something else. so what would be an example of that where you feel like there would be evidence you know the question I think people would have is okay like literally you're going to have an app that can solve you know diabetes or solve depression like is that is that is that possible like what would be what would be an example of where this might happen and where there would actually be you know potentially clinical evidence or

reasons to believe that it might actually work I think that's the real question because I think it's very natural for someone to be skeptical in this area and you know how could just an app or social networking actually have this impact? And actually, when I talk to friends of mine who actually are clinicians, and I talk about some of the work that my lab or my companies have been involved in drug design, they will often, like, put their arm around my shoulder in a very friendly way and say, you know, it's not that we necessarily need more drugs. I wish people would just listen to what I told them to do if they would just eat better and sleep and exercise and take care of themselves. And it actually is amazing how much that would actually have a huge impact on human health. And diabetes is a great example of that. It's, you could take a drug to help your diabetes or you could control what you eat.

You could do exercise. You could do all these other things. And this is type two diabetes. Type two. Yeah, yeah. Behavioral, largely. Behaviorally determined. Yeah, that's an important distinction.

Yeah, absolutely right. Type two diabetes. And so towards that end, the problem, the reason why the pill is appealing is that it's actually hard to actually do all the things that you're supposed to do. It's not that anyone wants to get type two diabetes and they're actively trying to do things. But in these lifestyle areas, that's where an app could, and a network could make a huge difference. And this is what we mean by a digital therapeutic, something where it can take the lifestyle

issues that need to be done and really enforce and help you along the way towards complying towards those ends. And in fact, we have a portfolio company that does this for diabetes called Omada. What level of evidence do you believe Omada has? Like, what level of evidence can Omata present either to users or to insurance companies or to employers that it can actually move the needle on something like diabetes? Yeah, you know, it's natural, especially for me coming to a science background that science is extremely evidence-based, and so we want to see, actually, does this work?

And so just like you could go to the evidence to see how well a drug works, you can actually compare a placebo, in this case, maybe just talking broadly about diabetes, to the leading drug to the digital therapeutic, and compare the effect in terms of how well the patient's doing in response to diabetes. And I believe actually Modis had done this. And in this case, one can show actually that the digital therapeutic is not just comparable to the drug, but in, and generally can exceed the capabilities of the drug. Yeah.

So one of the things you mentioned is we talk about digital therapeutics, you talked about the role of social networks. Maybe, I don't think people really necessarily wrap their head around this. Social networks are still viewed as something primarily that are, you know, most people, I think now think social networks are fun and they're enjoyable, and you're on Facebook and you're on Twitter and Instagram and you're having a good time. You're sort of hinting at a role for social networks and actually improving people's health. Can you talk about, like, what does that mean?

Yeah, that's a great question. And I think we think about social networks in terms of communication, but a lot of, I think, what it is to handle these challenging issues is to be able to have the inspiration and the, you know, you could pose it as inspiration which is a positive thing or sort of as a carrot or stick, as someone sort of on you all the time to make sure you're doing this. So actually. And exercise, better diet, less smoking, less drinking. All of these things, I think, are in these categories. and having your friends or people that are a part of a team associated with the therapeutic there sort of to have your back. That makes a huge difference.

I think it's just the way human nature works. Another thing I find fascinating is at least my outside interpretation of the field of medicine is medicine is very comfortable with there is a problem and we are now going to fix it, which has, and there's been tremendous progress across a broad range of medical issues over the decades because of that. It seems like more and more if you look at long-term health care spending in the U.S. U.S. or if you think about the health problems that we all are likely to have, especially as we get older, it seems like more and more they're not just things that happen to you. There are things that happen as a consequence of things you do, kind of as we've been discussing. So, you know, you don't exercise, you don't eat right. You don't do this. You don't get enough sleep. And things will go wrong. And as we solve more and more of the things that can be solved with a pill, the things that can't just be solved with the pill that are kind of lifestyle determined become a bigger and bigger percentage. And I've seen estimates that, you know, as much as something like 70s. of long-term health spending in the U.S. is going to be correlated to issues that are

caused, at least in part, by behavior. And so it seems like, you know, sort of paradoxically, it seems like the next big set of advances in medicine might be really helping people deal with behavioral things. And at least my interpretation is in the field of medicine, these are viewed as fuzzier issues or at least harder issues to get at, but it feels like they'll be rising much more to the forefront. I think so. And that's actually what's, that's both the interesting opportunity and the challenge here,

because if you're bleeding, we know what to do, or if you have a bacterial infection, we know what to do. And here is trying to take care of these issues before they become more serious. In the context of diabetes, there is this medical class of being pre-diabetic, which is something of a concern. Or maybe you have high cholesterol, or there's tons of markers that you may get early on to say that there will be a problem. And what's intriguing is that in the case of, let's say, high cholesterol,

all, if you don't do something about it in your 40s, in your 50s, the medical implications are quite significant. But with all these things, which is intriguing, is that you actually can also get feedback to see if it's working. You get the initial indication that there's a problem. Instead of taking a drug for one of these things, it would be intriguing to use one of these digital therapeutics. And it's not just something where you sort of roll the dice and hope, you can actually

see the progress. That's digital therapeutics. That's one category, or sort of example of this new generation. of bio companies. What would be another category that you think is interesting? Yeah, a second category that's interesting is sort of lutes to what we're talking about earlier with the building up of the analog

to fabs for semiconductor or for cloud computing. And this is an area that we've been calling cloud biology. And this is not running calculations on AWS that involve biology. These are biology experiments that are done in a cloud-like

way. And what makes a cloud-like is instead of one company owning one lab, much like it used to be that one company owned one data center. Now there's a shared lab that's shared by many, many companies. And you might run one experiment on Tuesday and a radically different experiment on Thursday. But the key thing is that like cloud computing, you pay for what you use along the way, and so you don't have to pay for the infrastructure. But what's even more exciting is that you can scale up and scale down. The elasticity of AWS is extremely appealing, the ability to run a calculation on 10,000 servers for the next 20 minutes and then go

back to three, depending on load. That's something that even if you built up a data center, you maybe don't even have 10,000 service in your data center. You maybe have 100 to handle the average load. Similarly, here, the idea that you can do experiments at scale and elasticity is really exciting. It's both efficient, but also in many cases you might just never have the lab that could do these things.

It also can dramatically improve reproducibility, right? That's a really great point, is that the reproducibility is something that I think is really underestimated. it comes up all the time that a large fraction of biology experiments are just irreproducible. Like numbers I've heard of shockingly high, like 30% or something. Yeah, extremely high. And at times, even 50, yeah, yeah, higher. Describe the problem for people who haven't heard of it.

Yeah, so the problem is that, you know, some experiment gets done and gets published and actually there's... So you're doing something, experiment with mice, let's say, and you're giving the drugs and observing behavior. And you publish your statistics, and it's done all very carefully in principle. And then actually, a second group goes to repeat the experiment. And they cannot be able to repeat it. And actually, there is even a journal whose sole go it is to repeat high-profile experiments because those are the most important to know whether they are reproducible or not. And if you think about how biology is done, there's a lot of challenges.

And it's, you know, done by really brilliant people. But there's just the challenges that a lot of the process involves a lot of human labor, a lot of people working their ass off at 1 a.m. to get this thing done to be able to really push science forward. And this isn't really nationally the conditions of all this doing this by hand under difficult circumstances is the best thing for reproducibility or for all these issues that we're talking about. Let's pause for a second. So I think if you describe this the average person in the street and if you said a research group got $2 million in funding and did an experiment and published an experiment and it can't be reproduced, most people would say, oh, that sounds like fraud. Yeah. Like, that sounds like somebody was being deliberately deceptive. You're saying that's not typically what you're seeing.

It's not typically the case. It's something else. Yeah. And actually, the thing about fraud in science is that reputation is everything in science. And so if one's experiments are reproducible, that alone, you know, is a type of thing that's extremely threatening to your career. And so there's all this psychological and cultural reasons to avoid this. But yet, even still, this is still a challenge. And I think that speaks to, not that people don't know what they're doing or that they have other motivations.

It's just it's a huge challenge. Well, and you have this sort of awkward thing, situation where you have these very, you know, brilliant scientists who are doing their scientific work. And then also, you know, administering a test on rats or mice, which is a very different activity. Yeah, that's right. Doing it under pressure. Yes. you know, doing it, you know, I don't know, all sorts of, it sounds like a different set of skills. It's a different set of skills.

Three in the morning. Three in the morning. You think about like just the industrial revolution is maybe not a bad analogy here, which is that when everything is manufactured by hand, that there'll be greater error rates and harder consistency machines just are better at consistency. Example here is Emerald Therapeutics, which is very much like this too and very code-driven. Reproducibility is built in. It's reproducible the way code is reproducibly. you just rerun it. And, you know, even in terms of where the robot is broken, these things, they run diagnostics all the time. So you even know about those types of things. So people who have been in biology for a long time would probably say, you know, they might say in response to this is, but there have been these things called contract research organizations for a long time. You have been able to outsource biology experiments for a long time. I think this is really fundamentally different. You know, you could outsource calculations before AWS as well. But the difference between running calculations on those older systems in AWS was, is the sort of essentially lack of friction that you have with ADWAS for spinning things up and down.

CROs are great, and they can, they serve a really important purpose right now, especially in the whole bio and pharma industry. But the amount of friction between a CRO and what you can get with cloud biology is dramatic. I mean, you're literally, if I understand it, you're on the foam with them, you're training them on procedures. Yes. You're doing what you would do with the freelance labor force. Yes. And as opposed to...

Whereas in cloud biology, you're literally writing code. You're literally writing code. Right. Yeah, exactly. Okay. Okay, good. So two great examples. Let's do a third. Yeah, so the third one, I think that's really something that is also dramatically changing is in the area of what I would call computational medicine. Doctors right now are sort of flooded with data, whether we're talking about radiology, genomics, all a series of tests. In each of these cases,

it's just a huge amount of information that's really getting to the point where it's beyond what an individual human being can incorporate and understand. even just on a single test, the resolution now that you can do on some of these imaging is so great that it's starting to even surpass what you could look at very carefully with the human eye. It's all these challenges, challenges of handling the flood. And this connects to all these beautiful current trends on the computer side of data science and machine learning. And computers have become extremely good at handling huge amount of data and for finding small patterns within the noise. and it's getting to a point where computers are exponentially increasing in their capabilities here

and people's capabilities have been flat over the last few thousand years. I think the exciting thing here is that it's not something where the computer is going to be the doctor or put the doctor out of business. Just like we use word processors rather than sort of writing by hand or typing, they're a huge aid to us to make us more productive. I think these tools will make the doctor considerably more productive because you can just do so much more. What would you say, like if you ask some people, in biology

and if you make these claims you're making now to some people in biology they'll say oh we heard this before we heard this in the 80s we heard the computers were going to change everything

and yeah we use computers and they're nice and email's great but fundamentally it hasn't transformed our industry you know what you say

it's different this time I don't you know I'm skeptical yeah no that's very natural I think because things are always different this time until they're different

you know and so And we've seen this in other areas. I think you look at the success of big companies that are driven by data like Google and Facebook. They've been very successful because of what they can do on the computer side, and they're integrating a huge amount of data. And so I think that's part of it. But actually, maybe what's special now is a confluence of that with many other trends.

There's the fact that Moore's Law has been chugging away and continuing to make a huge impact and getting to the point now where we can do computations routinely that we just. just essentially used to be impossible. But, you know, this is a special time in other areas. Genomics is making it such that we can sequence your human genome or your biome or a cancer tumor and do that cheaply and routinely. And then finally, I think, and this is relevant on the digital therapeutic side, too, the ability to have all these sensors, whether it's sensors in your phone or sensors

that are sort of being involved in the experiments in cloud biology, the cost of sensors going to zero connected with mobile, is also a unique opportunity right now. So, yeah, I think, you know, we can only prove that this is something that has happened until after it's happened, but there is something very special about what's going on right now, and we see all these parts coming together. Yeah. So a question about genomics, the role of genomics and the genome. So, you know, I think if we were sitting here, I don't know, 15 years ago or 12 years ago, there was huge excitement around decoding the human genome, sequence of the human genome, sequence of the human genome, sequence of the human genome. genome. And it was this giant project. And, you know, there were very well-respected experts in the field who at the time, I think, predicted, you know, a revolution in

cures of, you know, for cancer and all kinds of things, you know, that was sort of quickly follow. I think if we were sitting here five years ago or three years ago, we would probably say that that had been a bust and that while we decoded the genome and found out that it actually didn't give us many new options in terms of treatments or drugs, where do you think stand with genomics today? Yeah, you know, I think this probably follows the very classic Gartner hype curve in that there's for any given technology, there's going to be a point where people are excited beyond maybe where the technology is. And then the back part of the hype curve is not as exciting as the peak, but is where things become real. And so I think

we're way past the sort of the huge amount of original enthusiasm and we're starting to see it become real. There was a huge amount of money that went into the original human genome project and that makes it possible to have cheap genomic sequencing now. And so it wasn't that one But in perspective, like it cost billions 15 years ago. And it's now like $1,000 now, $40 to sequence soon. But if we didn't get huge results out of the first one that was very expensive, why do we think we'll get better results out of a million that are much cheaper? Yeah, that's a great question.

And I think it was a misunderstanding for maybe how the genome could be best used. I think the way people are viewing it today is that the genome is useful to tell us about how we're all different from each other. and this has such wide-ranging implications. It could tell us which drug to give you versus someone else that's radically different. It could even be in areas of cancer, cancer tumors are constantly changing. And knowing what drug to give is an extremely difficult question. And the ability to sequence a cancer tumor and use that information to tell you which drug to take is extremely dramatic. One example I loved is that I heard recently from an entrepreneur is that there are, you know,

better than I, but hundreds of seemingly effective treatments for cancer. But the big problem is knowing which one to use in which cases. And there's been a lot of progress, companies like Foundation Medicine, in building technologies to do that matching. And it turns out, because of the drop in these, in the cost for sequencing and all of the kind of physical aspects of this process, that it now becomes, in many ways, a software problem. So, so, In some ways, we have 200 cures for cancer. We don't know which ones to use. And figuring that out is a software problem.

Is that right? That's exactly right. And I think this is very clear on the cancer side where there's a huge number of drugs. And it's never going to be that the cure to cancer looks like a single drug. It could be sort of like with AIDS, like it's a cocktail or whatever, a mixture of drugs. But even at least in AIDS, it's a single virus. Cancer is really an umbrella term for many, many, many different. Having a drug to cure cancer is like a drug to cure disease.

You know, so it's just a mistake and a kind of a misunderstanding. Misunderstanding really how heterogeneous cancer is. The step one is identifying the type of cancer. Yeah. And is identifying the cancer and then figuring out the initial drug. But the part that actually is a surprise is that initial drugs can fail to work, and that's scary. But actually the upside is that drugs that didn't work at early stages will work at later stages. And so now the challenge is how to figure out which drug to give at what stage and what time. That would be impossible without genomics and without understanding where that tumor is.

And so it's intriguing that, you know, there's so much of us that are different and that those differences change in time, whether we're talking about our cancer tumors changing or even, you know, I think we forget that so much of our bodies is actually just by cells is not human cells, but bacterial cells in our gut. And the microbiome in our gut is also constantly changing. and so being able to understand those changes at a molecular level now is possible which is extremely sci-fi kind of stuff and now has become quite routine but the routine part is getting the data

the not routine part is what do we do with the data and that becomes I mean what you have in these things is you have sort of this dramatic change when it goes from $1,000 to test your stomach bacteria to close to zero dollars right then you can do it all the time you can continuously monitor it You just get a whole new set of things, activities you can do, right?

Yeah, that's exactly right. Across these kind of magic thresholds, and the same way we have with, in computing, we've seen this happen with storage and processing, like, storage is effectively free. And that means you can store all your photos online and you can do, you know, all these other wonderful things we do on the Internet and social networks, et cetera. You know, there's a big, big, big difference between close to free and $100 in this case, right? Yeah, yeah, yeah, exactly. Yeah.

So let's talk about economics, the implications of, what Chris is talking about in terms of how startups in biology work. So I think if you say, you know, bio startup or life sciences startup to an experience VC, part of the look of horror that you'll get is the idea of major regulatory hurdles. And, you know, we've talked in the past about there's, you know, the IT side of startups, the computer science side of startups is driven by Moore's Law, which is this, you know, incredibly fast exponential decline in the cost of chips. You know, there's this concept on the bio side of something called E-Rooms Law,

which is literally Moore's law reversed instead of more, it's Eroom's law, which is, and I don't know exactly what it is, but the price of getting FDA approval, literally the ticket price, the cash required to get FDA approval for a new drug or a new medical device has skyrocketed over time. And for a lot of new drugs now, the price I think is in the billions of dollars, which is way beyond what a startup can do. People call it Eroom's law, which is Moore's Law backwards, because it's followed the opposite pattern of Moore's law, which is...

Yeah, right. And so biostartops in pharma and in medical devices now have become terrifying to investors from a capital requirement standpoint. And in fact, a lot of new biotech startups get bought by big drug companies early now because it's only the big pharma companies that can really foot that. Or they spend hundreds of millions of dollars from VCs, then go public to try to raise more money to try to get through the FDA process, right? And then may or may not get that money. Yes, right. And it's just a, it's just a really, really expensive endeavor. Yeah. So these new, you know, we went through the three categories of,

you know, we went through sort of digital therapeutics, we went through cloud biology, and and we went through new kinds of computational medicine as our three examples. So are these new kinds of biostartops? Are these also subject to Eroom's law? Yeah, that's a great question. And, you know, what is the sort of interesting aspect of these is that they don't have the same profile in terms of FDA regulation. Digital therapeutics is not coming up with a small molecule drug,

and so therefore it wouldn't be something that would go through typical phase clinical trials. And in each of these cases, it doesn't have the same type of exercise. exposure. And so that actually has the hope to be fundamentally different. Combined with the fact that it has its own Moore's law, it has the Moore's law for computer, for genomics. And so while Eum's law is exponentially increasing cost for drugs, the fact that what drives all of these areas are essentially software and computation and or genomics, and those costs are exponentially decreasing, we would expect that there should be a radically different behavior. So these new startups have the potential to have the kind of economic profile and the kind of financing needs of a software startup. as compared to a pharma startup? Yeah, you know, if you think about it, these new startups, you know, remind me a lot of software startups in 2005, when we're starting to see the cloud computing

and other things start to realize, that's sort of what we're starting to see now. And because they have software at sort of their heart, either literally or in terms of how they think about things, that they're organizing themselves in a sort of cloud-like biology way or so on, this would be very much on the Moore's Law curve of things. And in a sense, you could use this to differentiate traditional biotech from this new crop of companies.

That traditional biotech is governed by Oom's Law, and these are governed much more by Moore's Law. If you started at dot com in 1999, it was probably an initial outlay, an internet company in 1999. It's probably an initial outlay of $20 million to get going, and you had to write big checks to these big companies like Son and EMC and Cisco and Oracle

to just get a website up and running. You know, kind of as you allude to, by 2005, and certainly by 2008-2009, cloud computing and AWS and open source and all these trends in computer science it evolved to the point where you have this boom of angel funding and seed funding starting new companies not for $20 million, but as an example, Facebook got started on $500,000. And then you had other examples, you have other examples of very successful companies since then

where the initial seed funding has been $100,000 or $50,000. And in the extreme case, you know, now new internet startups get started and it's, you know, three kids in an apartment, you know, in their entire KAPX budget is their laptops. Yeah. And whatever supply of ramen noodles they need for nine months, right? And it's like, you know, it's like you can, like, start a new internet company and, like, run an experiment with a new product for literally hundreds of dollars. And you go on AWS and light up and go global and see if it works. And, like, you're out, you can put the whole thing into your credit card.

So do you think, will bio, you think this new form of computer science-driven bio will get all the way there? In other words, is it possible that we will see an explosion of experimental bio startup? Yeah, well, we're already starting to see that. I think numerous companies that are coming out of places like Stanford or all over a place where it's, graduate students who have grown up with the biology and grown up with computer science, they've got all the intellectual tools they need, and they see AWS and they see emerald or they see a mouse sarah, and their brains are just clicking that basically half a million to a million dollars is all they really need to get the job done. And that could get you something through pre-clinical, which is what mouse studies are,

such that you could have something go into phase one or phase two with that initial seed funding. And that's dramatically different. Normally it would be like, you know, series. D or past that or post-IPO before you even get into sort of later stage clinical trials. Right, right. So the consequence could be, we could, we have the potential to see an explosion. Yes. Sort of seed stage bio companies running experiments in all kinds of different areas. The failure rate might be high, but it doesn't matter because you'd run so many more experiments.

You get a lot more successes coming out of it. Yeah, that's exactly right. And I think that's what makes software so appealing is that you can fail very quickly and understand. And for a low cost. And here I think we'll see exactly the same thing. Okay. Great. Let's maybe, let's spend the last few minutes.

talking about you. So maybe if you could give us a thumbnail sketch of your of your background. And then I've got a few questions about the ventures that you've been involved in. Yeah, sure. So I've spent the last 15 years at Stanford, but before then and during then, I've been an entrepreneur. So actually, you know, my first involvement with computers was when I was 12. Actually, we moved from Long Island to a suburb of Washington, D.C. And that summer, my parents did two things which was dangerous. We moved to a new place where I had no friends and I spent the summer with a computer. It's all downhill from there. It's all downhill from there. So I was in my first startup and I was 15 and this was a naughty dog, a software, a computer game company. And

that was actually a lot of fun and exciting to be involved at an early stage. And, you know, so my initial love of programming started to become lucrative, which was also nice, especially it doesn't take that much money to make a 15-year-old happy and startups can do such things. So then I went to college and I studied science, mainly physics, but with a biology, a sort of twist to it. And on the academic side, my work has had sort of interfaces with biology and chemistry and computer science and so on. And at Stanford, I... You're also a PhD is in physics. Yeah, my PhD is in physics, but actually I'm at Stanford.

I am in the chemistry department primarily, but also in the structural biology and computer science and a chair of biophysics. So sort of the whole sort of buffet. of all these things. And, but, you know, it was interesting when I got to Stanford within, I think literally within weeks, various venture companies asked me to evaluate companies, venture firms asked me about their companies. And this is not an uncommon thing. And it quickly made me introduced to these companies and I became advisors to them.

And that was, it's just part of the fun thing about living here and being connected in the ecosystem. And did you realize that before you got her? No, actually, I did not. I was at Berkeley for four years before then, and at that time, the two were actually very different now. I think things are starting to bleed into there as well. But that was actually really exciting to me because what often happens is that you do basic research and people ask, well, that's great, but can you impact disease?

You learn about the disease. They say, that's great, but can you make a drug? You know, you make a drug and it's like, that's great, but can you get it into the hands of patients? And at each stage, I felt the sort of push towards being able to make a drug. impact greater, greater. And having these connections with companies is the way to have that impact. Got it. And then there's two big projects you've been involved in. I'd love you to describe. So one is folding at home? Yeah. Yeah. So my group founded folding at home in October of 2000. So now we're actually almost up to its 15th anniversary. Wow.

So folding at home gets people throughout the world to donate computer time. And I think a lot of people don't realize this that in calculations involved in biology and chemistry, computer time could still be very much the rate limiting factor that in order to do calculations that we've needed, it would normally take even on the most powerful supercomputers decades to hundreds of years to get these things done.

So that's the first surprise, how much computer power is needed. The second surprise is just how much computer power is sitting all around us. The most powerful supercomputer in the United States for science is about at the 20-pet-flop level. So that's 20 times 10 to the 15th power. So 15-0. Of operations per second?

Of operations per second, yeah, exactly. You know, if you think about it's like, if you think a person with a calculator is one operation per second. Right. And you have a billion, let's say, 10 billion people in the world, which is an overestimate. You know, it would still take, it would still take 10 to the four, so 10,000 worlds. Right. Of people with all those calculators to do that type of, to that type of... To do what a supercomputer can do.

Yeah, yeah. But, you know, 20 petaflops is really quite small. And another way I think about is that, like, each... GPU puts out right now, a modern GPU easily puts out a terraflop. And a GPU is a card in your PC that you... Use for games and graphics. You're playing Call of Duty.

Yeah, yeah. It's a GPU that's putting all those pretty pictures up on the screen. Yeah, exactly. But it does a ton of math. And so if we just put a million GPUs together, that gets us to a thousand petaflops or an exoflop, which is like this huge holy grail of computing. You know, that's only a million GPUs.

You know, there's like a billion GPUs on the planet. right so there's just so much compute power that goes underused you have 400,000 now on folding at home yeah so we have about 40 petaflops and as a mixture of 400,000 computers and a whole bunch of GPUs and these are just people who are sitting at their desk at home or having fun or experimenting or find this stuff interesting yes and they download an app onto their PC or under their phone and it contributes into this kind of global supercomputer that's exactly right and we have the algorithms to be able to get these guys connected it's fun by the way because you get to see that it's like a cool picture on your screen I do it for that reason

So what does folding at home do, like what is the thing that it does? Yeah, so we're trying to understand the fundamental process for how proteins is part of the body, how they work. And when proteins work well, they do all these great things. They act as enzymes. They allow biology to happen. But when they don't assemble correctly, you get diseases like Alzheimer's or cancer or several other things. And a lot of these diseases are very difficult to understand experimentally through just test tube experiments. And much like, you know, in other cases, when we design car,

we design bridges, we will simulate them first before going into the lab and doing something. Here, if computers could be powerful enough, we could simulate and do things that you couldn't do in the lab and they make very specific suggestions. And having done these simulations, we can actually have made predictions for various drugs and drugs for various different diseases.

And with those in mind, then it usually sort of, I take off the academic hat because we want to push it forward and then it goes sort of more into the company space. And that's where we've been able to have, I think, sort of impact in both areas. And so for people listening, if you want to see pretty pictures in your screensaver and contribute to

solving disease in the world, you download, you go to the Folding at Home website, you download the app. Folding.com. Folding.combe.combe.combe. Folding.combe.combe.combe at home, it was sort of inspired by Seddy at Home. Yeah, very much so. And so DeVee Anderson's work with SETI, I think, was a great inspiration for many of us.

I think the challenge was... The Debt Home is search for extraterrestrial life, looking through radio signals from outer space. Yeah, exactly. The challenge is that... Which they haven't, it hasn't yet

been successful. Yeah, not yet. They've had better or worse. False positives, right? They've had, they've had things pop up that turn out to be the microwave oven down the hall. But, and I think the challenge, and this is a challenge much like any sort of thing in cloud computing,

how do you use like a million processors, you know, efficiently due a calculation? So these are challenge, sort of traditional computer science challenges in the modern era. Got it. Okay. And then let's, let's wrap us. talk about Globivir. Yeah.

And I'm really fascinated about what Globivir is doing and maybe describe kind of your role and then describe what the mission of the company is and how it's going. Yeah, so Globivir is a company involved in infectious disease and areas related to it. And I think our, you know, the challenge ahead of us is that there are many areas of infectious disease that are a huge nightmare for us that we don't have drugs for our therapies for. You know, we had this crisis reasonably about Ebola and that's something which was a huge disaster in Africa, but a lot of scare here. or two. But, you know, there's other examples. Dengue fever is something that kills 100,000

people a year in Asia. And actually, one of my relatives, this is something that's on their mind, the ones that live in India. Another area that we've been interested in Globivir is also the area Shagas disease. Chagas disease is the top five killer in Latin America, but due to global warming and other and immigration and other things, these insects are moving north. And this is becoming a serious issue. MPR called it the new HIV in America. And so, for all these things, these are infectious disease where we don't have existing drugs, but our interest has been to use computational methods

to repurpose drugs, to be able to find new purposes for existing drugs. And this is something where it allows us to move very quickly. In this sense, it's very much in the spirit of these other companies we've been talking about. It's computational heavy, being data-driven, using that data in intelligent ways, using algorithms,

and then moving in ways that can sort of, sort of accelerate the regulatory process. So instead of taking 15 years and $100 million to do the first part, we could, in the case of our drugs for a lot of these areas, we could do it in nine months. So the drug's already been shown to be safe. Yeah. And now you need to show it's effective for something that it wasn't originally designed,

or tested for rather. Exactly. And I think part of the surprise here is I think most people's knee-jerk reaction is, well, how could something be useful? How could a drug be useful for two things? like isn't it hard enough to be useful for one thing and I think the reason why this is such a surprise is that the brand names given to drugs

masks the chemical names and so there's many drugs that actually are the exact same thing but are for use for different indications like the drug for in Unisom is the same drug in Benadry and it's actually it's masked by the fact that it's different chemical names and this is actually

there's many examples of this and so you know so the question isn't could a drug serve multiple purposes. That's actually been empirically shown in many, many cases. The question is, could you identify what drug can do the purpose you want? And that, like a lot of these data questions, is a very natural computational question. And so we're able to address that and then move very quickly. And I think what I like about this approach is sort of how computation really come in and make an impact, but also how we can sort of do our best to try to speed things

through the normal regulatory environment. Great. Well, we're at our time. So thank you very much, VJ, and this will be the first of many on these very exciting topics. Yeah, great. Thank you.