Gregory Bowman on How You Can Help Cure the Covid-19 From Home

Transcript

Harry Glorikian: Hello, I’m Harry Glorikian. And this is Moneyball medicine. The show where we meet executives, entrepreneurs, physicians, and scientists using the power of data to reinvent healthcare from machine learning to genomics, to personalized medicine. We look at the biggest trends in patient care and healthcare management.

And we talk to people behind the trends to find out where data is making the biggest difference.

With the COVID 19 pandemic wreaking havoc all over the country. We have developed the utmost respect for our healthcare workers who are fighting on the front line, protecting us and saving lives. However, there is another group of people behind the scenes who deserve just as much respect for their work and those people are the researchers.

Tirelessly working every day and night to find a cure for the Coronavirus while not all of us are medically trained to be working on the front line. What if I told you now that you have an opportunity to be a researcher and potentially contribute to the discovery of a cure for the Coronavirus? Our next guest is the current director of Folding@home Dr. Gregory Bowman, who is also a professor at the Washington University in St. Louis. And he will be shedding some light on what he does, how he does it and what we could do to help accelerate this process of finding a cure. Dr. Bowman, welcome to the show. It’s great to have you here. 

Gregory Bowman: Thanks. It’s a pleasure.

Harry Glorikian: So Dr. Bowman, tell us a little bit about yourself and a little bit about the project you’re working on.

Gregory Bowman: Yeah, I have a really long standing interest in biomedical research having lost most of my vision to a juvenile form of macular degeneration as a child. And about the time that the implications of that started to dawn on me, uh, it was right as the human genome was being sequenced and Dolly, the sheep was being cloned.

Uh, and so I got really excited about the potential to be a part of this effort, to understand complex biological systems like ourselves and how we can control and fix them. Uh, so I I’ve been, uh, involved in this track for a long time {inaudible} your own computational biophysics degree as an undergraduate student at Cornell?

And then we’re off to graduate work at Stanford. And now I’m on the, uh, stint at Berkeley as a fellow. And then on the faculty at the Washington University school of medicine now where continuing to work on understanding protein dynamics and implications for drug design, for example.

Harry Glorikian: So you’re working on particularly, you know, Folding@home. And so if you were to sort of explain Folding@home to your grandmother, or a child, like how would you frame it so that everybody understands what Folding@home is and what it, what it does?

Gregory Bowman: Yeah. Great question. So Folding@home is basically a big distributed computing cluster with the primary focus of understanding all the moving components of a protein and how they contribute to its function and malfunction and diseases like Alzheimer’s disease and cancer, uh, and how it can use this information to design, do therapeutics. And ultimately one of our major objectives is to use the now millions of computers around the world that are volunteering to run simulations for us, uh, using all of this, to build a map of the different structures that have protein adopts and how the protein can hop between them.

Uh, and the map analogy is a good one. We can’t, we can’t just take a satellite image of this space that we’re trying to explore. So essentially what we’re doing with all of our volunteers is sending them out like cars that are driving around (our} local vicinity and sending us back the GPS coordinates. And on our side, we piece this all together into a map that extends far beyond the reach of what any individual explorer did.

Harry Glorikian: So basically you’re just using the power of everybody else’s computer to supercharge the work that you’re doing.

Gregory Bowman: That’s right. Exactly.

Harry Glorikian: So how would you put that into some sort of measurement that someone could get their head around? In other words, how would you compare that to you know how we would think about spinning up systems on Amazon?

You know, how would you make a comparison?

Gregory Bowman: There’s a couple of different perspectives one can take. So, so one is the peak performance in terms of number of computations we perform for a second. So our estimate right now is that with all the volunteers that we pool together, we are the first computer to reach the excess scale.

Uh, so that’s a unit of measure. Uh, currently the world’s fastest supercomputers are measured in petaFLOPS instead of exaFLOPS. 

Harry Glorikian: Yep

Gregory Bowman: So, so our estimate is that we, our peak performance is about 10 fold that of the world’s fastest traditional supercomputer. So this is an enormous computational resource.

Well, even, even before the a hundred fold growth that we’ve experienced and starting our work on COVID-19, we were routinely running calculations that would have cost millions of dollars to run on the cloud, you know? So, so now a hundred fold increase in compute power. We’re talking about hundreds of millions of dollars or more worth of simulations.

Harry Glorikian: You got to use these systems when people make them available. So does that change the time it takes to do this?

Gregory Bowman: Yeah. So it’s a really interesting paradigm, right? Because normally on a traditional supercomputer one is thinking about how do you get all these tightly integrated computer working together on what is essentially one big calculation,

Harry Glorikian: Right?

Gregory Bowman: And our whole approach is how do you take this one big calculation? And break it up into pieces that can be performed completely independently of one another with no communication between these computers. Uh, and so this has been really fascinating because it allows a tremendous level of scaling. Uh, and we get a bunch of features, like fault tolerance for free.

So if someone turns off their machine or their heart or whatever, slows us down a tiny bit, but we just keep crunching along. Whereas uh  failures like that, can be very hard to deal with in traditional supercomputing setting.

Harry Glorikian: Right. Right. Right. So before the COVID 19 pandemic, so what were you guys working on?

Gregory Bowman: Yeah, so we’ve, we’ve got a number of research labs that are involved in Folding@home. And each of us is tackling a number of problems. So prior to the pandemic, you know, my group alone was doing a lot with Alzheimer’s disease and the Ebola virus and antibiotic resistant infections. Each of which can be thought of as a global pandemic in its own rights.

Uh, we’ve done a lot with cancer, uh, collectively as well, multiple forms of cancer. So there’s quite a broad range of biomedical problems, as well as very basic questions. Like, the original project that fully focused on which was how to proteins fold into these functional structures so that the breadth is really a fun part of this and that the generality or this platform.

But for the time being, we’ve put basically everything else on hold and we’re bringing everything we’ve got to bare on COVID-19.

Harry Glorikian: Using COVID-19 is an example, sort of explain the science behind Folding@home and what you’re specifically trying to do with, with COVID-19.

Gregory Bowman: So one of the challenges at the length scales involved with proteins is that we can’t just zoom in on them with really powerful microscope and watch what’s happening on these scales.

Everything is very indirect. And so one of the consequences of this from an experimental perspective is that we can usually at best build a model of what a protein usually looks like, right? And this is immensely valuable information, but it’s also far from complete. So, so taking American football as an example, you know, if you’re asked, what are the players usually look like? It’s them lined up at the line and scrimmage, you know, 

Harry Glorikian: Right

Gregory Bowman: Waiting for the next play to start and there’s a lot of information there, but really what you want to see is the game unfold, right. To see who actually gonna win and you know what strategies should you employ? Um, So, what we’re doing with our computer simulations is filling in all of these moving parts and seeing what we can learn from that, about how a protein functions or malfunctions and how one could get in there with drug design for example, to uh fix problems. So, so with COVID-19 as a specific example, when you see these pictures of the virus with all of these often red protrusions sticking off of the surface, uh, each of those is a complex of three proteins, called the spike, and you always see them in those images, in this closed conformation where they’re packed up against each other.

And one of the really cool things here is that one of the ways the virus {inaudible} and a new response is that in this closed state, the interface of that set of proteins that actually binds to a human cell and initiates infection is closed up and protected from being recognized by the immune system.And so for the virus to infect us, this thing has to open up

Harry Glorikian: Right

Gregory Bowman:  Like the mouth of some organ monster uh, and in order to expose that surface. And make it accessible to engage with a human cell and, and little is known about the open state or how this process of opening occurs. And we’re now able to capture that with Folding@home, uh, and start exploring whether the different stages of this motion might serve as valuable drug targets for example. And, and going beyond the spike. Now we’re looking at basically every protein from the virus that we can build a reasonable starting model for hunting, for moving parts like this that might be interesting drug targets.

Harry Glorikian: So if you were to describe some, maybe in more detail, like what a proposed solution is that you might be able to bring to the table with the technology that you’ve got. How would you describe that to someone?

Gregory Bowman: We’ve got a bunch of things going in parallel right now. And this is one of the really exciting things is that we’ve very much switched from a mindset of what, what should we do to make the best use of this limited resource to ah, what else should we be doing?

Harry Glorikian: Right

Gregory Bowman: There’s so much breadth of compute power. Uh, so we’ve got a number of things going. One of the things that has the greatest potential for impact on a short timescale is that we’re performing enormous computational screens of large libraries of chemical compounds, uh, searching for those that are likely to bind most tightly to some of the key components of the virus.

Uh, like the main protease, which is one of the uh, popular targets for drug discovery and then through a project called the COVID Moonshot. Uh, we’re teamed up with a number of experimental groups who are taking those predictions of all of the tens of thousands of compounds we’ve looked at computationally.

What subset should we actually buy or synthesise and test experimentally to see if it binds and inhibits this important target? So, so that’s one of the things and then where we’re hunting for these alternative structures of these proteins. As I mentioned with the spike or demogorgon, as we like to call it and starting to take those essentially novel structures that no one has seen before and gain insights into how these things work and also use them as starting points for a rational drug design. Uh, and we’re, we’re starting to explore antibody design as well. So, uh, our strategy is try everything we can in parallel and we’ll, we’ll see what sticks.

Harry Glorikian: So that was going to be one of my other questions, which is like, you’re producing all this information, I guess who’s consuming what you’re producing that would then be able to make a therapeutic against it, or are you trying to do both?

Gregory Bowman: Great question yeah, so, so we are a publicly funded operation. Most of, most of our funding comes from, uh, the NSF and NIH. And so we try to do things as in the open as we can. Uh, so in the past, we’ve always shared data sets upon requests after publication.

Uh, in this case with the immediacy of the pandemic, we’re making extra efforts to share our data as quickly and as broadly as possible. So we’re engaged with a number of cloud service providers and are working on ways to put these data sets, which are pretty large,

Harry Glorikian: Yeah

Gregory Bowman: Up on the internet for anyone who wants to dig into. And so our hope is that in much the same way as, uh, having more computers work on these problems in parallel, uh, accelerates progress that, you know, making the data available and allowing more brains to crunch on it and see what value we can extract will also accelerate progress.

Harry Glorikian: Yeah. AWS has now a, um, a data exchange platform that they’ve created for scientific data.

Gregory Bowman: Yeah, that’s right. Yeah. So we’re talking with their public datasets team.

Harry Glorikian: One of my friends outruns it. So he’s been showing me sort of what they make available, which is interesting of how they want to play that role of intermediary or providing a system for everybody being able to exchange data.

Gregory Bowman: Yep. So that’s one of the projects and there’s, there’s a couple others that are also trying to help with this, which is great because we and others are producing enormous amounts of data, it’s not trivial to just put it up on our website, for example.

Harry Glorikian: Yep. Uh, well, and there there’s, you know, there’s the production of the data and then there’s the analysis of it, which sort of, you know, brings me to like, so you said you’re focused on the demogorgon, right? The, uh, the stranger things, 

Gregory Bowman: That’s right

Harry Glorikian: A beast. Um, but. Is that the, is that the protein that you’re really focusing on and have you started to understand its structure and mechanism of action or where you think something can be targeted against it?

Gregory Bowman: Yeah, so the spike complex and the main protease are some of our initial focal points.

Harry Glorikian: Okay

Gregory Bowman: 28 proteins and the viral genome. And we have simulations up and running then about half of them at this point, as well as different complexes between them. Uh, so, so our hope is to, you know, on the analysis side, focus our mental energy on a subset of them starting (with) the spike and the main protease, uh, and to get the data out there on the others for others to chew on as well. Uh, and we’ve made making a lot of good progress. So, so this past week, uh, the first batch of experimental tests, uh, based on predictions coming out of Folding@home were performed on the main protease. And there’s some, some interesting leads in there that weren’t, uh, further follow-ups. So that’s a really exciting step to have done the experiments in addition to, you know, finding, uh, tantalizing hints that, uh, more progress is in the works.

And, uh, and we’ve been finding what we call it, cryptic pockets in the main protease. Uh, so these are really fascinating because, you know, often when you look, at the experimental structures of one of these proteins, there are a few, or maybe no concavities or pockets in the surface for a small molecule drug is likely to be able to bind tightly enough to serve as a candidate or a drug development.

Uh, but often when we go and look at all the moving parts and what they’re doing, we see, uh, these cryptic pockets that are absent

Harry Glorikian: Yeah

Gregory Bowman: In that {inaudible} open up, and we can often find that they actually would be quite valuable, uh, targets for drug discovery. So we’re making progress on that front and we’ve also, uh, already, uh, you know, written some blog posts and tweets about actually seeing the opening up of this spike complex and starting to understand what that looks like and what opportunities for things like antibody design that might open up.

Harry Glorikian: Yeah. I mean, in one of my previous podcasts, I interviewed Jake Glanville about Distributed bio. How there, you know, Engineering the old SARS antibodies to work against coronavirus. And I think I just saw that they were tweeting or talking about having some strong success where they’ve had four different laboratories test out the antibody against the spike protein itself. 

Gregory Bowman: I’ve enjoyed hearing some about their work.

Harry Glorikian: Yeah, the sooner we come up with something to manage this disease, the better off, I think everybody will be. But when we talk about designing drugs against some of these proteins and why it’s so difficult is can you sort of explain that just a little bit? I think it’s difficult for people to understand that if you’ve got the key, you need the hole where it can go in and unlock something and sometimes that hole is not available.

So how would you describe that if you were walking somebody through why something may or may not work well?

Gregory Bowman: Yeah, {inaudible}  really- {inaudible} one of the challenges of working at the scale, which is that the macroscopic scales that we typically think about life seems very deterministic and, uh, maybe it’s a little too strong to say predictable, but there’s a lot of predictable aspects of our life, right?

Uh, at these small scales, everything is very stochastic and very {inaudible}. Right. So, so if you see something happen once it doesn’t mean that’s usually what happens. You could have just seen an outlier, uh, in fact there’s many different outliers. So you have to observe things many times to get a real sense of what typical behavior looks like.

Uh, and, and further complicating things. Everything is strongly coupled together. Right. So if you’re stuck at home on quarantine, you know, you know that if you open your front door, that is independent of what’s happening to your backdoor or your windows, for example, but then in the realm of proteins, that’s not true opening the front door might cause the backdoor to become impossible to open essentially, or it might cause it to spontaneously spring open and your windows to go up and down. And so it’s really hard to predict what the effects of small changes, like adding an atom to a drug like molecule are going to be, uh, because there’s all kinds of, uh, responses that come into play that are often at odds with each other and make it very difficult to predict exactly what’s going to happen.

So there’s a huge amount of trial and error in the drug discovery process. One of the things that we ultimately want to do with building a home is try to reduce that by having these quantitatively predictive models that lets you make much more accurate guesses out, how the system is going to respond to these small changes.

Harry Glorikian: So essentially the more and more people that jump onto your bandwagon, the faster and more effective you’ll be able to be able to look at these proteins and run simulations and come up with opportunities.

Gregory Bowman: That’s right. That’s right. Now we have a pretty insatiable thirst for more compute power and scientifically I would happily put all the world’s computers to work {inaudible}.

I’m sure I would still run into things where like, ah, only I have a little bit more we could do and such and such, you know, one of the really exciting things now is that the combination of technologies, you know, including the computer hardware and the scientific code for simulating these things have all in recent years really come together that we can do a lot of really useful stuff that was completely impossible when I was a graduate student, you know, and or, and certainly before that. So, you know, it’s a really exciting time where computers have, uh, an unprecedented opportunity to accelerate biological and biomedical research, as well as other fields.

Harry Glorikian: Oh, yeah, there isn’t a day that goes by where I’m like, I need to go back to school.I mean, most of this stuff wasn’t even like, I, you know, I see high school students doing things that we weren’t doing when we were in college and I’m like, geez. Wow. That’s, that’s a totally different type of experiment. So. Uh, makes me feel like I always threatened. I tell my family, I’m like, I got to go back and get another degree.

Like, it’s just, just so I can go and play with it myself. But, so if you were to sort of walk somebody through the process of implementing this or, or doing this, how, how would you describe it to them since we don’t have any screenshots or anything like that, per se?

Gregory Bowman: Uh, so the process of like setting up and running one of these simulations {inaudible} ?

Harry Glorikian: Yeah. I mean, if they were sort of to set it up on their end and what, what do you see on your backend, I guess?

Gregory Bowman: Right. So, so what we’re doing on our side is taking these, you know, experimental structures that give us just one snapshot, what this really complicated thing full of moving parts looks like, and we are, uh, surrounding it with water and other aspects of the solvent and environment these things live in, and now we’re taking this atomically detailed representation where in very simplistic terms you can think of there being a little skier for every Adam and protein, and they’re hooked up with Springs to represent ions and they’re pushing and pulling each other based on whether they’re positively and negatively charged.

So we hand this off to these molecular dynamics algorithms that are over and over and over and over and over again, asking, you know, given, given the ways these atoms are pushing and pulling on each other, where is each atom in this system going to be some small time in the future. And one of the reasons that this is so computationally expensive is that small time is a really small time, like 10 to the negative 15 uh, seconds. So, so, and, and this is because, you know, if we, if we tried to work on longer timescales, we would get unphysical behavior like atoms jumping through each other, 

Harry Glorikian: Right

Gregory Bowman: Which just doesn’t make any sense in order to watch these tiny things collide and bounce off each other. Uh, we really need to work at these very, very short timescales.

And then we need to do billions and thousands of billions and millions of billions of repetitions of this to build up to the timescales that are relevant to biological systems and human life, you know, where enzymes, catalyze reactions and proteins change structures in response to light or odorants and all the other processes that we’re interested in.

Uh, so in our side, you know, each of these, thousands of simulations that we might run over protein onFolding@home, it’s kind of like a movie, right?

Harry Glorikian: Right

 

Gregory Bowman:  As a human interaction, you could start off in the same conditions and, and every time you press play, you know, something a little bit different might happen.

And now what we can do is take all of this data and build a map of the space of possibility. Again, returning to this analogy. You know, taking GPS data from lots of cars and using it to build up a map of this really large space that no individual car may have seen all of them. 

Harry Glorikian: Right. Right. So, so how do you see, you know, the evolution of Folding@home?

Where do you see it going in the next, you know, year, three, five, et cetera.

Gregory Bowman: Oh man. Um, I’m super excited about the folding, uh, future of Folding@home now, as long as we’re still here and, uh, uh, you know, so, so the project started almost 20 years ago now. And, uh, the initial focus was really on this very basic research question.

How do many proteins start off with this extended linear chain of chemicals called amino acids, and like fold up on themselves into these reasonably specific three-dimensional structures that can perform these amazing functions like catalyzing uh, enzymatic reactions better than anything that we’ve come up with the {inaudible}, for example. Uh, and since then the project has really evolved to broaden that scope. So our initial steps were starting to think about misfolding diseases, protein misfolding diseases, like Alzheimer’s disease and Parkinson’s and Huntington’s disease. Uh, and then we started to grasp what these moving parts are really essential to the functions of many proteins then we started. So we started thinking about, Oh, you know, how do these things change in response to mutations implicated in cancer or Alzheimer’s disease? And what can we learn from that about how these things work and how we could develop small molecules to fix them. And so looking ahead, um, um, yeah, really excited that we do have this very general platform that you essentially can be used to understand how anything that small and made out of atoms works uh, which is quite a broad reach. So the, the initial focus will continue to be on different scales and organization and biological systems. But yeah, looking ahead, we also have the opportunity to think about how can we draw on ideas from deep learning to analyze all this data.

Harry Glorikian: Right 

Gregory Bowman: And how can we apply these tools to material science? Uh, so, I think things are super exciting.

Harry Glorikian: Yeah, no, I mean, I’ve talked to a couple people in material sciences and it’s sort of interesting, uh, working on, uh, crystal structures and being able to sort of get an idea of what net, what the next molecule should look like before you’ve even made it, the capabilities that are coming online.

It almost seems daily because I can’t keep up with the publications that are coming out, fundamentally going to change the way we do science.

Gregory Bowman: No, definitely. I mean, there’s a lot more opportunity for. Comparing and contrasting, contrasting things and trying lots of things. Right. So, so I mentioned that, you know, we’re, we’ve got simulations up and running of at least half of the proteins, I think from uh, SARS CoV 2. And, uh, no, but it’s not just those like 12 or 14 proteins. We’re also got simulations up and running of the, uh, variants of these proteins from the original SARS virus. So we can ask, well, why is it that SARS 2, it’s so much of a bigger problem for humanity right now. What insight does that give into how to deal with this and comparing to other Corona viruses that have been around for a while, but not cause nearly as much of a headache to understand what has changed.

And this gets really exciting then also for considering large numbers of possibilities for drug and antibody design, where we can, you know, have, have some intuition guiding things, but also try lots of parallel after the fact asked, Which, which of these trials that we ran, uh, looks most fruitful?

Harry Glorikian: Right, right. Well, it’s interesting, right? I mean, I think to myself that, you know, not that long ago, it was sort of 80% lab and 20% computer capability or, or IT, and now it feels like it’s the reverse, where it’s 80% computational capability and 20%, you know, lab sort of work.

Gregory Bowman: Yeah. I mean, and certainly the opportunity to combine them is one of the really exciting things.

So my lab at this point is a third or half experimental, and we spend a lot of time hopping back and forth between these detailed insights. That would be very hard to get by any other means besides our computer simulations. And then designing experiments to test these insights 

Harry Glorikian: Right

Gregory Bowman: And to test our ability to get in and manipulate things with small molecule changes to the chemical composition of proteins, for example.

Harry Glorikian: So, I mean, based on where you are, I mean, I’m, I’m sort of seeing a fundamental shift needed in the person doing the work. Right. There’s, there’s gotta be an understanding on the comp side as well as the biological side to coming together. Um, and I, I’m not sure we have graduated enough people in that combined domain yet.

Gregory Bowman: Yeah. That’s definitely a challenge just that, you know, within my lab, for example, we’ve got. Yeah, deep things going on with biology and chemistry and physics and machine learning and artificial intelligence and computer science and information theory. Uh, so I think one of the really exciting things is that we get to go and explore all these different regions of human knowledge, but it definitely also poses the challenge that it’s, uh, uh, a challenge to get sufficient grasp with all of these things, but this is one of the really interesting things with, uh, you know, is it makes it more, it’s more important to work together in teams.

People need enough awareness, and enough.uh, breadth to help appreciate the opportunities, uh, but then to go deep in their different sub-domains and be ready and willing to, uh, work as a team.

Harry Glorikian: Yeah, no, I think that, you know, where we see the next breakthrough is this sort of cross-functional teamwork coming together um, but I also think that labs like yours are great places to go and find people to hire

Gregory Bowman: That’s right.

Harry Glorikian: Well, I mean, I can only wish you incredible success, uh, you know, the sooner we come up with a way to tackle COVID, um, I guess we’re just, maybe we should ask people to jump onto Folding@home and, you know, put their computer available so that you can get that one more online to help you do more and more work.

Gregory Bowman: That’s right. That’s right. Every little bit helps. And, uh, you know, one of the messages I’m always trying to get out there is that Folding@home is uh, about more than just, uh, COVID certainly this is the most important problem immediately, but looking ahead, we do have this very general platform and I hope, uh, people that jump on board with helping with our COVID 19 work well recognize the value of applying the same approach to Alzheimer’s disease and Ebola virus and all of these other problems and stick around with us and help us prioritize those in the future.

Harry Glorikian: That’s a good question. Now that you brought that up, how do people know that their contribution is making a difference?

Gregory Bowman: Yeah, we do a lot to engage with our community. So we, we publish scientific papers and we also write blog posts that try to be more accessible to the general public. And we’re very active on social media.

So Twitter and Facebook, for example, uh, and we have a number of active communities where people can get engaged. So we have a forum @foldingforum.org. That’s mostly focused on helping people with technical issues related to participating in Folding@home. And we recently started a discord server, which, uh, people can go with the technical questions too, but that’s also more of a forum where people can, you know, discuss with each other and with members of our scientific team to try to understand what’s going on and what opportunities are out there?

Harry Glorikian: No, that’s a, that’s a great way to extend someone’s already existing education or just a, you know, knowledge

Gregory Bowman: That’s right. that’s right.  It’s a, you know, really fun STEM opportunity that people are very hungry, especially now and understand what the principles behind our work are and what we’re doing and what the results are looking like and the potential implications, uh, uh, really, really exciting to see those levels of engagement and interest.

Harry Glorikian: Well, excellent. I wish you incredible success, uh, with what you’re doing. And like I said, the faster and sooner that we get there, the better 

Gregory Bowman: That’s. Right. Thanks so much. 

Harry Glorikian:Thank you. 

Gregory Bowman:Thanks to all our volunteers. It’s been tremendously helpful.

Harry Glorikian: And that’s it for this episode, if you enjoyed Moneyball medicine. Please head over to iTunes to subscribe, rate, and leave a review. It is greatly appreciated. Hope you join us next time until then farewell.