Leadership Insights: Open Source GROMACS Clinical Research

Transcript

Mike Vizard: Hello and welcome to the latest edition of the Digital CxO Insight Leadership Series. I’m your host, Mike Vizard. Today we’re with Erik Lindahl, who is a professor at Stockholm University and is also a member of the Royal Institute of Technology. And we’re talking about how drugs will be developed using open source-type technologies and protocols. Erik, welcome to the show.

Erik Lindahl: It’s great to be here, Mike.

Mike Vizard: So you guys are working on this project, I think it’s called GROMACS? I mean, I’m not sure I got the right pronunciation there. But what exactly are you guys trying to do?

Erik Lindahl: So, this is an old problem. If you’re Dutch, the form of pronunciation would be GROMACS, because this was originally developed in Groningen, in the north of Netherlands, but we’ll call it GROMACS, nowadays. This is based on the idea that all molecules in your body, proteins, everything, DNA, they, really, they consist of small atoms. So this goes all the way back to Richard Feynman, that in principle, if we just know all the equations behind this, we should be able to predict how very large molecules such as antibodies, viruses, DNA really work in practice. Which would be a miracle for modern drug design. Now, this has remained a miracle long-outstanding promise for decades, but over the last 50 years, due to this exponential growth of computing power, that means that we can now start to simulate things that are not just on physical but chemical and even biological scales.

We can simulate how the proteins in our bodies fold, we can simulate how the spike protein in COVID, how it actually moves and how it’s binding things. And we can simulate how small molecules such as pharmaceutical drugs occasionally bind to these molecules. Which, of course, holds great promise for using it to screening, so move things from the lab to the computer _____.

Mike Vizard: And you’re getting support from Intel and others, so, what exactly is the IT community helping you with, here?

Erik Lindahl: Yeah, so the problem is, 50 years ago, it was easy to program a computer: you just wrote your program in Fortran, few hundred lines of code, and then it was the compiler and vendor’s problem, and then everything just worked. And it was that way for 20-30 years, but, roughly, when I was a student, we started having these parallel computers, first. That was fairly trivial; we managed to get that to work. But computers have become increasingly complex, and I would even say a single chip, today, is much more complex than the largest supercomputer we had in Stockholm when I started.

And in addition to these simple chips with 32, 64, or even more cores, we now had so-called accelerators. If you’re a consumer, you think of them as GPUs, but to me, a GPU is not really a graphics device; it’s a massive device with some 5 or 10,000 very simple functional cores. And our challenge is, how do we learn to make this pony, well, not do graphics, but do simulations of our pharmaceutical molecules. And that’s a gigantic undertaking where we’re working very close with Intel, to make sure that these codes will run the next generation of hardware, in particular, Intel’s.

Mike Vizard: Are these simulations themselves gonna be open source software? Will they be shared in the industry? And how will that change the way –

Erik Lindahl: [Crosstalk] we are not – it’s a bit strange, I care a lot about the value of software, and I am actually a very big fan when it comes to IT and the pharmaceutical industry, for that’s the reason we exist. But these codes are complicated, they are really, really, really complicated. The problem is that we make one small error, say, in the physics, here, our simulations are not gonna predict the right thing. So in terms of the scientific correctness, here, it has tremendous value for us that are not just ten pairs of eyes in my team, but there are thousands of pairs of eyes all over the world trying to find errors. And we’re pointing out each other’s mistakes, in a very friendly way, and that’s, of course, how science improves. And I would argue that all the major code in this fields, just as bioinformatics, are open source for this reason: it makes the science go faster.

Now, when it comes to the actual simulation, it depends. A whole lot of people, including my team, in my own team, I’m not necessarily developing pharmaceutical drugs. I’m a professor of biophysics, so I wanna understand, fundamentally, how these molecules move. That is, why does a particular virus infect your cells, or why does a particular channel in your body collect nerve signals. And those we publish; they are completely open. And we are beginning to see a change, here. Twenty years ago, even the pharma, the pharma industry was quite worried and they wanted to protect their IP. But I think here, too, they are increasingly defining a precompetitive area, that is, the early-stage research that is more focused on understanding, where they happily share, because they, too, are pooling knowledge the same way. But, of course, at some point, it does switch over to competitive research, and then they’re using all these tools to basically develop their pharmaceutical IP, and that’s the way it should.

Mike Vizard: So it’s kind of like the same way we think about open source software, where there’s a lot of underlying components that aren’t differentiated value, that we don’t all need to rebuild the same thing over and over again. But then there are things we add value on top.

Erik Lindahl: Exactly. And because we all pool into that, the base layer, because we’re all pooling our knowledge there, right, that means that we all get ten times more value in the base layer than if we all tried to do it competitively. But I think it works just as well in industry as in the academia, in my case, or precompetitive part of pharma.

Mike Vizard: So, do you think this model will find its way into other vertical industries? ‘Cause it sounds like there’s a lot of places, now, where we’re trying to use simulations, and it’s the same kind of issues.

Erik Lindahl: I think I would argue that it already has. So now, of course, now you’re talking about fields where I’m not necessarily the expert, but bioinformatics, it’s very hard to develop anything commercial in bioinformatics today, because the open source software is so much better. If you’re looking at fields such as people simulating our brains or so, everything is open source. Computational fluid dynamics, there, I would argue there might be a bit of a bifurcation, there. The large academic codes, the ones that run the supercomputers, they’re all open source. And then there are some codes that are more focused on desktops today, they tend to be commercial, mostly, not so much based on the value of the code, but the value in the support.

And here, too, I expect that we’re gonna see some sort of convergence in the future where the codes are free, but of course, if you want somebody to hold your hand, you’re gonna need to pay for the support. And that is the case in our field, too. There are several companies out there that use open source tools, or rather, they support open source tools, they offer commercial support, that hand to hold, for a company who would like to do this but they don’t have the expertise inhouse.

Mike Vizard: One of the challenges, historically, has been that we could only really do research on areas where there was gonna be a large return, so, most of the focus was on diseases or issues that affected a large number of people. Do you think that this is making it more economically viable to go after cures for diseases that affect a smaller number of people, because the cost of doing the research is coming down?

Erik Lindahl: Yes and no, maybe not the computational aspect, per se. So first I would argue, this is true if we’re looking in terms of money, but what one should be aware of that the early stages of drug design and pharmaceutical research are relatively cheap. The cost goes up exponentially once you actually wanna go into the clinical tests or so. And that, of course, means that it is easier to find a commercial market for something, so, the major disease in the Western world. But that early preclinical research is happening, it’s happening in academia, even in Sweden that, due to our climate, we do not have malarias, but there are researchers performing academic research in malaria, across the road from me, here.

And I would argue open source or, say, open data, which today goes beyond software, right, everything in machine learning, the data is frequently more valuable than the actual algorithm, we are seeing a trend where everyone is sharing data all over the world. And that is enabling researchers in, say, the Third World and everything where they might not have the resources to develop gigantic wet labs, but it’s quite possible for a small group to work on the same datasets as the large groups in, say, the US or Sweden. So there is, I wouldn’t call it a democratization aspect, but I think it’s, yes, sharing both data and algorithms and ideas definitely drives both science short-term but also business long-term, because many of these drugs will be commercial successes.

Mike Vizard: Do you think it’ll become possible for research teams in smaller universities that don’t have access to a lot of horsepower, locally, to use the cloud and use these types of services to participate in this research, and we may see more people doing this kind of research?

Erik Lindahl: It’s already happening. So first, computers have an advantage that you don’t need to sit right next to the computer to use it, right? So, all over the world, US, Europe, Japan, China, people can, small research teams can get a small allocation on a computer, first. And if you show that you use it well, there will be more time for you. And that is so much easier than starting a large wet lab. Trust me, I’ve done both. But the other part that we’re seeing is that we are also having citizen scientists, occasionally, for several years, that both we _____ _____ have been involved in folding at home, right? And there we don’t – the individual contributors are not necessarily the scientists deciding what proteins to simulate, but they’re contributing a few cycles from their computers or graphics cards, every day.

And in terms of disseminating the value of science and making sure that people understand why we can do it, I think it has tremendous value. Now, second, as we’ve seen in COVID, these distributed networks enable us to collect orders of magnitude more data than we could do on the larger supercomputers. So it’s not really – I think it’s easy to think of it as toy science. It’s definitely not toy science. It’s one of the most valuable computational resources in the world, right now.

Mike Vizard: Do you think the pace of innovation and joint research, then, is going to accelerate? Are we on the cusp of some sort of renaissance period?

Erik Lindahl: I think we’re gonna see both. So first, computers are making everything go so much faster. The big shift we’re seeing is, of course, the trend towards bigdata, and bigdata, it is an overused buzzword. But the whole, I think pharmaceutical research is an example of – the problem I am solving is a fairly simple one, right: I wanna understand how one molecule binds another molecule. And it is important, but it’s just one very small part of pharmaceutical research. When it comes to actually making a drug, understanding how we can administer the drug, the metabolism, the excretion, the toxicity, there are hundreds of other factors to consider.

And all this data, we collect this data from various trials, and we know things about the solubilities of drugs and toxicity effects, and this is a miracle field for machine learning, of course. And the machine learning, so, a) computers accelerate this, but machine learning accelerated even more, so we’re definitely in a renaissance for pharmaceutical development.

Mike Vizard: And it is somewhat humbling [crosstalk].

Erik Lindahl: [Crosstalk] the companies wanna collaborate with academia, again.

Mike Vizard: And it is somewhat humbling that, despite all our advances, we still don’t really understand how, fundamentally, molecules interact with each other. And this is at the root cause of many of the issues that trouble us as a society, right?

Erik Lindahl: Yes, or rather, I’d say, when it comes to one molecule, we understand it. The problem is that there are many, many, many molecules in our bodies. So, if I know that it’s Molecule 14, here, I can understand. The problem is that, in practice, in a cell, there will be 10,000 other molecules. How will this molecule interact with the other 10,000 molecules? That I don’t know. How will the other 10,000 molecules interact with my receptor? That I don’t know.

And this complexity is still challenging and we haven’t solved that. But there are very – as much as we love the lab and there are many things we can in the lab that we can’t do in the computer, but on the lab side in my lab, our wet lab does not get twice as fast every year. And the computers do, so that this difference will keep growing.

Mike Vizard: So if somebody wanted to participate in this project, where would they go? How would they get started? How do they get in touch with you guys?

Erik Lindahl: So there are two parts to that. If you’re a citizen interested in science, get involved in folding at home. And I am not saying this primarily because you would donate resources to us; we do appreciate that. But I think it is a great project where we also have several of the PIs that will talk about their science, why are we doing this, explaining the projects. If you’re, on the other hand, say, an undergraduate student interested in doing this type of research, you can do almost anything. You can study physics or you can study computer science or you can study some parts of chemistry or medicine. And then, try to focus your studies on this somewhat ill-defined area right between these fields.

So that much of the hardest work in my teams comes from computer scientists, but of course, we couldn’t do what we do without the physicists defining the equations, and the biologists or MDs actually defining what are the specific molecules we need to work on. So I kind of, I’m not quite sure what area we’re in anymore, but we enjoy it. And as long as the science is good, I don’t mind what we call it.

Mike Vizard: Is there a shortage of expertise in this space? It seems like there’s a shortage everywhere [crosstalk] so what’s different here, anything?

Erik Lindahl: So then, one big challenge is, of course, that we’re – I’m primarily looking for people who are both physicists and computer scientists and experience from the wet lab. They rarely exist. And people skilled with computers and – and again, we tend to use buzzwords today, but high-performance computing, in the words’ purest sense: computing where performance matters. That can be machine learning, it can be bigdata, it can be traditional massive simulations on supercomputers, but it’s not only us. There are a hundred more groups in the world, and every single large computer _____ center, both in Europe and the US, they are – they could hire ten more of them instantly. And I could likely hire three or four.

Mike Vizard: All right. Hey, Erik, thanks for being on the show.

Erik Lindahl: It was great to meet you.

Mike Vizard: All right, and thank you all for watching this episode. You can find this one and all the other ones we’ve done on digitalcxo.com. We invite you to check them all out. And once again, thank you all for spending some time with us.