Transforming Drug Discovery: Inside GSK's Data-Driven R&D Revolution

[00:00:02] Today, I'm joined by Dr. Christopher Austin, SVP of Research Technologies at GSK, and together we're going to unpack how technology is revolutionising the way that we develop medicines. And Chris has somewhat of a unique background as a medical doctor, neurologist, biotech entrepreneur, and someone with experience across academia, government, and the private sector.

[00:00:27] And at GSK, he's leading the charge in leveraging AI and machine learning to accelerate drug discovery and ultimately improve patient outcomes. So today, I want to explore how can AI and genetic validation improve drug success rates, and understand why generative AI is also revolutionising disease modelling, drug design, and clinical trial simulation,

[00:00:53] a little bit more than write a blog post or email for you, and also explore what the future of AI-powered medicine could look like in the years ahead. So, the intersection of technology and medicine is reshaping the pharmaceutical industry, we're told, and also changing lives. But how will AI shape the next generation of treatment? Well, let's get my guest onto the podcast to find out more. So, thank you for joining me on the podcast today, Chris.

[00:01:22] Can you tell everyone listening a little about who you are and what you do? Sure. Well, it's great to be with you. So, I'm a medical doctor by training, and that really drives everything I do. I'm now the Senior Vice President of Research Technologies at GlaxoSmithKline, and it's a large multinational pharma company based in the UK. And I thought, how does a medical doctor end up focusing so much on technology? And it's really an illustrative answer.

[00:01:51] Not only a medical doctor, but a neurologist by training. And when I was in training 40 years ago, it was really evident that the vast majority of patients that I saw, I could do nothing for. We didn't understand their illnesses. We didn't understand what was causing them. And we certainly didn't have any treatment for them. And I would go through entire days in the clinic without having a single patient that I could help.

[00:02:17] And it was clear that we needed game-changing ways to understand and treat disease. And those leaps only come from technology. And so, in my own case, I started with human genetics in the era of the Human Genome Project. I was involved with many new technologies, all of which have been developed during my professional lifetime, incredibly, in the last 40 years,

[00:02:41] including genomics, robotics, gene therapy, oligonucleotized small molecule chemistry, things called human-induced pluripotent stem cells, data science, AI. And I've done that across academia, big pharma, spent a lot of time in the government at NIH, ran a startup biotech for a number of years, and now big pharma.

[00:03:03] And I believe, and this is why I'm at GSK, that big pharma, in particular GSK, is perhaps counterintuitively the place to be for the next big gains that are possible. And I'm really excited about what we're doing at GSK to, as we like to say, bring science, technology, and talent together to get ahead of disease. And it's that proactive rather than reactive approach to healthcare that feels incredibly exciting.

[00:03:32] And I think it's only been the last four or five years that we've seen the actual impact that technology can have on healthcare, in particular biotech. And now we've got artificial intelligence and machine learning and so much more going on. And we hear a lot about this in our news feeds. But how are you seeing technology transforming the traditional drug discovery process and, indeed, the development process at GSK? Anything you can share around that? Yeah, sure.

[00:03:59] For those of probably a lot of your listeners may not be familiar with how the drug development process has traditionally worked. So the problem has always been an unrelenting empiricism in the therapy development process. That is, it's been virtually all trial and error.

[00:04:19] And the drug development process from a discovery in the lab to getting a drug developed and deployed to all the patients who could benefit from it is roughly 20 steps. It currently takes about 30 years to go from beginning to end with a low success rate at every step. And increasing, like, degrees of freedom or difficulty to have that process, each of which has a low success rate over a long period of time.

[00:04:47] The likelihood of your getting from the beginning to the end with a trial and error process is extremely low. And that's what we've seen on the order of 0.1% success rate from start to finish. And changing that requires how the human body works. And that has now become possible because the average researcher now generates about a million times as much data per person per time that was possible when I was in medical school 40 years ago.

[00:05:16] And thanks to computation and AI, we have an analytic capacity, which is far beyond what the human brain would ever be able to handle with all those data. And that analytical capacity is now matching that.

[00:05:31] And what's really exciting from this is that we've gone from recently, just as you say, in the last four to five years, from analyzing large amounts of data, which was a feat in itself, to summarizing those data to true reasoning. And that is a huge breakthrough. So the way I like to think about this is, traditionally, imagine yourself if you're trying to get from point A to point B, and you don't have a map.

[00:06:00] You know where you are, and you might drive around in fairly random directions, and eventually you might find what you're looking for. But now, for the first time, we are developing a comprehensive map. It's like a Google map of biology. And so if we want to go from point A to point B, we can predict it in ways that simply were never possible before.

[00:06:21] And that means that we're able to go faster, and we can reach our goal, which is a drug that dramatically improves health and longevity of people in ways that were simply impossible before. And of course, data tech, including AI, machine learning, and generative AI all plays a significant role now in understanding patients and disease mechanisms.

[00:06:44] But I'm curious, if we zoom out for a moment, how can these technologies ultimately enable GSK to accelerate drug discovery and that ultimate goal, of course, to improve patient outcomes? Anything you could share on that? Yeah. So what we mean by data tech, when we start, we just start at the beginning.

[00:07:03] If we're trying to understand what a patient actually has across hundreds of thousands or millions of patients, to understand the panoply of disease, of symptoms and signs that are independent of the diagnosis that they are given. Doctors are trained to give diagnoses, and those can be very practically useful. However, imagine where they came from.

[00:07:27] These are disease categories that were developed over the last hundreds of thousands of years, which tend to group people together who may have different mechanisms of whatever is wrong with them. And using large clinical databases and AI to be able to disentangle in individual groups of people, we can now target individual types of individual people and groups of people by the cause of their diseases, not just their symptoms and signs.

[00:07:56] It gives us biomarkers that we can follow. It gives us mechanisms. And all of those technologies then lead us through that map of human biology to choose the Achilles heel of a disease that we can target.

[00:08:10] We then use these same technologies or same type of technologies to design a drug, whether it's a small molecule or a protein or an antibody or an oligonucleotide, design those rather than the empirical trial and error method that we've used before. And then we can perform clinical trials in a much, much more effective way because we know the patients that we're looking for.

[00:08:36] And so we can choose them to have a high likelihood of responding to our drug given the symptoms and signs that we're looking for and the biomarkers that we're looking for. And I'll just give you an example. There's a drug called Depomocumab for asthma and a number of other respiratory disorders that GSK has developed.

[00:08:55] And using these technologies has allowed us to go from a phase one trial, that is the initial first in human trial, to a definitive phase three trial in a very short period of time. Actually for four different diseases, because we knew those four different diseases were all related to each other in having the same cause of the disease that was treated by Depomocumab. So it's an example of the kind of thing that's now possible.

[00:09:22] And target selection is critical to drug development. So how are you at GSK using both data and AI to prioritize genetic targets most likely to deliver those meaningful health impacts? Yeah. For those of you who may not think about this every day, a target, we use the word target to mean the molecular entity.

[00:09:46] It's usually a protein, but it could be another molecular entity in a cell that we design our drug against. And finding the right target in the right disease has actually been a major reason for drug development failure in the last several decades.

[00:10:03] So what human genetics now allows us to do is to look across the entire genome of large numbers of individuals and using both genome sequencing and a variety of other technology called omics, transcript omics, proteomics, metabolomics. We're able to identify in groups of people through their genetics.

[00:10:29] What is the particular genetic circuit that's gone wrong in these cells, in these patient cells? And by circuit, I mean a group of genes that all work together to do a certain thing. And when that certain function goes haywire, like a circuit going out in your house, the lights go off. And when the circuit goes wrong in a person, then we get disease.

[00:10:52] And it's always been hard to identify kind of like when your lights go out in your house, you have to go down to the circuit breaker and test each one individually to figure out why the lights went off. That's what target selection has been like. What we're able to do now is to identify without testing them, which one is wrong and where is the defect in that circuit.

[00:11:13] And that all comes from the development, not only development of human genetics, but now 25 years after the genome project was finished, we're beginning to understand how all these genes work together to cause disease.

[00:11:25] We're able to prospectively go from what is wrong to a patient to what is the genetics causing that dysfunction to what is the circuit that's abnormal, the multi-gene circuit that's abnormal to help reverse the disease.

[00:11:45] And all of that is only possible because of the large data generation and analytic and AI capabilities that we now have. The other thing I should mention, in case your readers aren't familiar with this, it's been shown originally by some investigators at GSK about a decade ago, that drug targets that have genetic evidence are at least twice as likely to succeed.

[00:12:15] So every target that we investigate at GSK now for drug development has that genetic validation. And that's a huge change from the past, particularly for common diseases. And if we dig a little bit deeper on that and geek out for a moment, when it comes to molecule design, how do you decide on the right platform of technologies, whether it be mRNA and small molecules or monoclonal antibodies to reach those specific genetic targets effectively?

[00:12:45] You know, when I was in training, and again, it's only about 40 years ago, not that long, there was one kind of drug. It was a so-called small molecule drug. These, they're relatively small organic compounds. And those were all drugs were those pretty much. Now, the meaning of drug has changed dramatically in the last 40 years. There are all kinds of different flavors of small molecule drugs.

[00:13:11] We have proteins, peptides, monoclonal antibodies, bispecifics, antibody drug conjugates, drugs called oligonucleotides that can degrade a target, can block the production of a target, can edit DNA or RNA. And of course, we can deliver full-length genes through gene therapy, and we can even deliver cells as in CAR-T or in microbiomes. So the question is not what is possible, but what is right for the right disease in the right patient?

[00:13:40] Very excited about RNA. You heard about mRNA during the pandemic, but these are smaller RNA or DNAs known as oligonucleotides. And we're very bullish on traditional and new kinds of small molecules as well. So how does one choose, given a target, which one to use? Well, really, Mother Nature decides that for us. There are some targets that are effectively reachable with small molecules.

[00:14:07] Other ones, such as oligonucleotides, can reach a large number of the potential therapeutic targets that are difficult to address using other technologies, small molecules or anybody. In fact, about 50% of therapeutic targets are only reachable, we think, via oligonucleotides. And that's why GSK is putting such an enormous effort in oligos. And I'll give you an example of what we can do with oligos.

[00:14:37] We are developing an oligonucleotide-based treatment for chronic hepatitis B, which is a global plague that causes liver cancer, cirrhosis, and death, a major cause of liver transplantation. And using oligonucleotides, we have the opportunity and entirety of that virus's genome.

[00:15:03] We can put in several different, six or seven different oligonucleotides that hit the virus in multiple different ways. And so there's no way that it can escape because we hit it in every possible way all at once. That's only possible with oligonucleotides. And patient identification also remains a key challenge in clinical research. So how are you using data tech to determine which patients could respond to treatments at various stages of their disease?

[00:15:34] Well, it gets back to what we talked about a little bit at the beginning. That is, we're using very large clinical databases to profile patients throughout the course of their disease so that we can understand what has gone wrong in individual patients and in groups of patients,

[00:15:54] starting with their genome sequences, but then looking at various biomarkers, which allow us to identify patients who are most likely to achieve clinical benefit. And as an example, in that example that I just gave you about hepatitis B, the drug is called Beppe Roverson.

[00:16:12] And we're using AIML to identify patients by their biochemical characteristics and the kind of virus floating around in their blood to achieve a functional cure in more patients with chronic hepatitis B. And not only does that allow us to target or treat patients more effectively, it also allows us to streamline our clinical trials because we're more likely to predict who's likely to respond.

[00:16:42] And if we think about what is the theme going through this, the theme is really predictivity. And if I go back to the oligonucleotide example just for a moment, traditionally, if we were looking for an oligonucleotide against a disease or against a particular target in a disease, we would need to test many, many, many of these in order to find one that works.

[00:17:09] Using AI now, we can look across the entire gene and predict which oligonucleotide, when oligonucleotides, as you just mentioned, are 20 subunits long, 20 building blocks long, out of a gene that could be several thousand building blocks long. So which one is going to be the most effective? We can now predict that via AI, which we were never able to do before. So again, increasing predictivity.

[00:17:40] Same issue with patient identification. Instead of taking a large heterogeneous population, whether it's with cancer or hepatitis B or asthma, defined on clinical benefits, clinical characteristics alone, we can look deeply into their genomes and into their genetic regulation, biofluids and biomarkers, and identify groups of patients are most likely to respond.

[00:18:07] And clinical trial effectiveness is also essential for timely drug development. And we've seen a lot of increases in speed in recent years. So how are you at GSK using technology to streamline those clinical trials? Anything you could share around them? I'll use the example of depomocumab again in this case. So this is an antibody against IL-5.

[00:18:28] We have extensive experience in that IL-5 pathway as interleukin-5, which is known to be diseases of a number of sorts. And by doing what's called deep phenotyping, that is looking in great detail at large numbers of patients' genomes,

[00:18:48] and as a result of that, we were able to go from a first-in-human study straight to a phase three study in four different indications simultaneously. And that allows us to design and carry out the trial in a much more effective, efficient way. One of the things that I have dealt with a lot in my career is that because we were not very good at identifying patients to go into clinical trials,

[00:19:18] patients were reluctant to take part. And clinical trials, upwards of three-quarters of clinical trials, would fail because they couldn't get enough patients to enroll in the trial. And that is now becoming a thing of the past because we're able to say to patients, look, this is not a random shot on goal. We're able to tell you this is the nature of your disease. These are the biomarkers that you have.

[00:19:45] And this is a drug that's very likely to treat what you have. And gosh, if patients hear that, they're eager to sign up. And that's what we're finding at GSK, too. It's dramatically accelerating our recruitment and the carrying out of our clinical trials. And there is a lot of hype around Gen AI in just about everything at the moment. And generative AI is also emerging as a powerful tool in research.

[00:20:12] So how do you see its role evolving at GSK around potential breakthroughs, especially around drug discovery and development? Because I would imagine there's a lot of promise, a lot of opportunities. We've got the hallucination problem thing going on. But how do you view it? Yeah. Well, Gen AI is still in relatively early days as an entity, an effective entity. It's only been around a few years. Generative AI, I mean.

[00:20:38] But we are at GSK aggressively incorporating all forms of AI, including increasingly generative AI, into everything we do.

[00:20:51] Everything from understanding and modeling these malfunctioning disease circuits that we talked about, to design of antibodies and oligonucleotides and small molecules in a much more efficient way,

[00:21:08] to modeling what goes on in a diseased cell through what are called digital twins, where we can reproduce an organoid that is a multicellular model of a patient given their cells, and then profile them over and over again to understand what's going on and then predict what the treatment is likely to work, all the way to simulating clinical trials.

[00:21:33] And before that, simulating what might happen in an animal, where we're looking for toxicity or safety, and then simulating that clinical trial as well. And in each case, what we're doing is taking large amounts of data and using those data, understand what the general principles are underlying that, to allow us to predict what's likely to come next. And I'll just give you one example, which I find incredibly exciting.

[00:22:02] All of your listeners will be familiar with antibody treatments, which have become quite common over the last couple of decades. Started out with a drug called Herceptin for breast cancer, but there are many, many others now.

[00:22:15] And traditionally, what that has required is to use animals that we would inject a particular antigen into, and we let the animal produce many, many, many antibodies, as the immune system likes to do. And then we would screen through those in a very laborious process to try to find an antibody which did what we wanted it to do.

[00:22:43] That's all gone at GSK. What we do now is we can use the knowledge of protein structure, tools like AlphaFold and others, to understand what it is about the building blocks of a protein that are going to fold into a three-dimensional structure, which we can predict is going to target whatever the antigen that we're interested in.

[00:23:09] And so a huge amount of the work that we do in developing antibodies now is all computational. We can go through rapid so-called design, make, test, model, design cycles where we design antibodies. We make thousands or tens of thousands at a time. We can characterize it with high-throughput capabilities. That then goes back into the design engine again.

[00:23:35] And within a few rounds and within a year, a year and a half, we're able to come up with a clinical candidate. That's one-tenth of what it used to take. And the amount of effort and the amount of people time and cost is drastically reduced as well. The same thing's happening with small molecule therapeutics, that design-make test cycle, all driven by design because we understand what the characteristics are that we're looking for.

[00:24:01] This is driven by very large data sets which have been developed over the last few years. Now, something that is really important and it's really the reason that I decided to leave biotech and come to GSK is the ability to create fit-for-purpose data. We've made a huge amount of progress with public sector data. And if we look at ChatGPT, it's a good example.

[00:24:27] That was all built on public sector language, written word out there. So that's why they're so effective, very large data sets. In biology, we tend not to have those. And it's become evident over the last few years that no matter how good the algorithms are, they get limited unless you have large fit-for-puts data sets that can drive the algorithms to increase productivity.

[00:24:53] The only place that can combine very sophisticated data tech, computation, AI, ML, generative AI, and the ability to generate enormous data sets fit-for-puts that drive the increased effectiveness and accuracy of those algorithms is a company like GSK.

[00:25:14] And GSK has made, in my view, the greatest investments into data and data analysis and data generation. And it's that opportunity that made me decide to leave running my own biotech to come to GSK.

[00:25:31] Because I am confident that we have the whole recipe that's required to really drive breakthrough therapies for untreatable diseases more rapidly than has ever been possible before. That's why GSK is such an exciting place to be now. It's just electric. And listening to your story today, it does seem that you're in somewhat of a unique position with vast experience across public, private, and government sectors.

[00:25:59] So from your experience and everything that you've seen, looking at GSK's approach to research technology, what is it that makes that stand out? You mentioned it's incredibly exciting there. Not only what makes it stand out, what lessons can the broader pharmaceutical industry learn from what you're doing here? Well, I think having been in all four of those sectors, I think the most is we're all trying to do the same thing. We're all trying to bring more treatments to more people more quickly.

[00:26:29] We tend to emphasize the differences. But for heaven's sake, we're all trying to do the same thing in complementary ways. So as in most complicated things, working together is absolutely essential and sharing data and sharing experience.

[00:26:42] What is so exciting about GSK is it is allowing me, as an exemplar perhaps of the whole field, to bring together decades of deep drug development experience with the most modern, up-to-date, and high-throughput data generation capacities with unprecedented computation and data tech capabilities.

[00:27:10] Those things have tended to exist in different sectors. Some have been good in the public sector or in government or in biotech or in pharma. But what I'm finding at GSK is all the individual things that I valued so much in bringing everything together that's required to develop a treatment for people is all present in GSK. Okay. So where are we likely to get?

[00:27:37] Well, it's now possible to do what I dreamed about 40 years ago. Given AI and given all the large data sets that we have, it's a Google map of how the human being works in health and disease, a full toolbox of different drug types to fix whatever's gone wrong, and using tech and AI to digitally design those drugs rather than trial and error.

[00:28:01] And the ability to identify what drug is good for what person, and therefore be able to design and carry out clinical trials in a much more effective way. So you might ask, well, gosh, we're not there yet. What do we need to get there? It is the continued development of new technologies, of course, and new computational AI algorithms. However, it gets back to the fundamental coin of the realm in science, which has always been high-quality data.

[00:28:29] And with our ability to drive unprecedentedly fast data generation, I'm confident those algorithms are going to continue to improve. So I think what I see, and that's going to result in better, faster, more effective drug development, think about the world that I've tried it for of my whole career, is to never have to tell a patient what I had to say so many times when I was in clinical practice.

[00:28:58] That is, I'm sorry, there's nothing I can do for you. That is crushing for patient, crushing for the doctor. Instead of that, doctors will be able to say, hey, what's the type of disease? And we can do that right now in these examples of hepatitis B, asthma, multiple other diseases that GSK is working on. This is the type of disease you have. This is what's causing it. And here's a drug that treats what you have.

[00:29:26] And it's becoming a reality now. I think that is a powerful moment to end on. We did start the podcast talking about your origin story, what put you on this path. And as we come full circle now, I'm going to ask you to look back and forget about the technology and the powerful work that you're doing there. Just for a moment, try and reflect on the funniest or most interesting story that has happened in your career that you are allowed to share with me today. I'm sure there's a few you can't, but is there anything you can?

[00:29:54] Yeah, I think it is a funny story. And it's a story, which is when I started out my career after I left academia and medical training and research training, I went to Merck or Merck Sharpe. And that's where I learned drug development. I had the opportunity in the early to go to the Human Genome Project.

[00:30:15] And the Human Genome Project, probably all of your listeners will remember, was an incredible technological tour de force of its time. Completely unprecedented. It was the AI of its time. And in going from a drug development organization to the Genome Project, I was considering to do that.

[00:30:43] The question that I got asked by the head of the Genome Project in the U.S., named Francis Collins, I'm sure you've all. He said, what you were doing at Merck on a few hundred genes at the time to try to figure out what these genes did that were just being sequenced every day. There were new sequences coming out of the Genome Project, genome sequences, trying to understand what I was trying to do at Merck and figure out what do these genes do biologically and what's their therapeutic potential.

[00:31:10] And Francis, his job offer to me was, well, how would you like to come down here and figure out what to do with the genome that is in understanding the biological function and therapeutic potential of all human genes? And I remember thinking at the time, hmm, how would you like to come down here and help us figure out what to do with the genome? I thought, wow, that is either a ridiculous proposition or it is the greatest job offer in the history of science.

[00:31:40] And if you remember that moment where the book of life was revealed to us and to be one of the people who had the privilege to understand what the genome did and design programs to elucidate it for the benefit of human health, that was a completely unexpected opportunity, but something from which I've never looked back. And it gave me an appreciation for a couple of things.

[00:32:06] First, it gave me an appreciation of the value of large data sets, right? I mean, the Human Genome Project produced the first large data sets that were publicly available on which we could model doing something called bioinformatics, if you remember that. It also taught me the importance of being bold, of being audacious in the goals that we have for patients.

[00:32:31] There are a lot of people who thought that the Genome Project was either impossible or would be irrelevant, but the people involved in it knew differently, and they set forward a really ambitious and visionary agenda by which technology could improve human health. And I learned a lot from that. The other thing I learned was that drug development is much more complicated than having a genome.

[00:32:57] So it requires a lot of other technologies and a lot of sectors to all work together. But that fundamental knowledge that we gathered there of how important large data sets are to be able to compute on, that is the legacy that led to the AI explosion that we're currently in. And it's taken longer than I hope to get to, but here we are.

[00:33:26] It's a very, very exciting time. And the patients and families that are listening to this, I hope you realize that this should give you hope that things really are moving in a direction where the kind of world that we want to live in is becoming a reality. Wow. What an incredible story. And we've covered so much in a short amount of time today. And many people listening, maybe they want to dig a little bit deeper.

[00:33:54] Just keep up to speed with the kind of work that you're doing. Contact you or your team, people in and out of the industry. Where would you like to point them? Well, the best place to go is gsk.com. There's a lot of information on gsk.com. I am an unabashed acolyte and booster of this too. So I'm on LinkedIn. I've always liked to hear from people. Glad to contact people that way too. Awesome.

[00:34:23] I'll make sure there's links to everything to make that process very easily. And one of the things I always say at the end of every podcast episode is technology works best when it brings people together. But one of the things I've learned about GSK today is how you're using technology to develop and deliver medicines and vaccines better and faster. But it's not about the technology. It's more about positively impacting the health of people at scale. So just a big thank you for sharing that with me today. Sure. Sure. It's been my pleasure. Enormous pleasure.

[00:34:53] Great to talk with you. From target selection to clinical trials, AI and data science are beginning to reshape the entire drug development process. Not just making it faster, but fundamentally more effective. And a few of the quick takeaways from my conversation with Chris today is how GSK are leveraging AI and genetic validation to increase success rate of new drugs. And in doing so, improving efficiency and precision.

[00:35:21] And I was also fascinated to hear more about generative AI, how it's already playing a major role in disease modeling, drug design and trial simulations, cutting down development time significantly. And with smarter patient selection, making clinical trials more efficient. I think ensuring the right people receive the right treatment sooner has to be a step in the right direction.

[00:35:46] But as AI continues to revolutionize medicine, what breakthroughs do you see that could be just around the corner? And how will these advancements impact the future of healthcare? Let's keep this conversation going. Please email me, techblogwriteroutlook.com, LinkedIn at Neil C. Hughes. Nice and easy to find. Let me know your thoughts on this one.