Ovation advisor and former Chief Data Officer at Foundation Medicine, Dr. Gaurav Singal, sat down with Ovation founder Dr. Barry Wark to discuss computational sciences, the history of genomics and RWD in Oncology and what’s next for the life sciences industry in TAs outside of Oncology.
Barry: Thanks so much for joining us today Gaurav, we’re super excited to hear about your experiences in the real world data and genomics space and talk about what the future looks like for life sciences researchers and patients. You’ve done a lot of work in these areas in the past – could you please share some of this background for readers who might not be familiar?
Gaurav: Thanks for having me Barry! I trained originally as a physician and computer scientist, but have spent the last decade working in cancer genomics trying to bring the power of large scale genomics data coupled with clinical data to have an impact on life sciences/Pharma R&D and clinical decision making at the point of care. That was an incredibly rewarding exercise – I got to both build a novel database and demonstrate its utility for research and regulatory purposes. I then also got to build a suite of applications on top of that platform, resulting in products and services that ultimately created real value for both clinicians and drug developers.
Barry: So what do you enjoy about being an advisor to companies in early stage growth like Ovation?
Gaurav: For the last several years, I’ve been advising a series of companies in the diagnostics, therapeutics, data, and nonprofit spaces. I have centered my work around three topics that I’m quite passionate about: advancing the impact of real world data; making clinical trials faster, better, and cheaper; and bringing computational diagnostics into the forefront of medicine.
I was really excited about Ovation before I even met you Barry. I spent a lot of time with Flatiron Health when I was at Foundation Medicine, where I helped develop and lead the partnership we had built with them. I was really intrigued by their model of providing an EMR to community oncologists, a customer segment that frankly is often neglected, and doing so in a way that they could harness real value from the data that was embedded in that software platform. I thought that was really creative and a way to create value for multiple stakeholders in a way that was accretive to everyone. With Ovation, my initial interest was that it had many of those same initial patterns. For a while I thought about and described Ovation as “Flatiron Health for LIMS” and that felt pretty interesting.
Ovation addressed another neglected sector, lab management software, that is not always optimized or designed well, and also has really interesting opportunities from a data perspective that could create real value for the world. Broadly, the RWD community hopes that the data they unlock could help develop medicines more effectively, with a higher probability of success, using more targeted patient populations – which we would all agree are truly accretive for the world. But especially in the independent lab universe, this data is lying fallow. That’s what initially attracted me to Ovation.
After formally joining and getting to spend some time with Barry it became clear to me that the ambition was actually dramatically greater than that. The newest push of Ovation to become, as I see it, a decentralized biobank, is incredibly exciting.
Barry: That’s great, that’s one of the best descriptions of our journey I’ve heard. Makes sense given the unique perspective you have on this industry from your experiences with Flatiron Health and Foundation Medicine and other other companies in this space.
Gaurav, could you talk a little about why you, and really the world, started with Oncology in the real world data space. Also what did we miss and what could we achieve if we take those learnings elsewhere?
Gaurav: Yes certainly, so let’s rewind the clock about a decade. I had just finished residency and my background was in computer science, artificial intelligence and machine learning. I had spent several years as a resident working on natural language processing for clinical care and for clinical research, but my real passion was bringing computational methods and computational science to the practice of medicine. I had spent some time at Third Rock Ventures so was in the orbit of Biotech but what I really was passionate about was bringing advanced computational data science techniques into the clinic. At that time there were a ton of challenges with that including:
- Lack of broad availability of data required for training and implementing data science solutions
- Lack of high quality methods, with relevant mapping to relevant use cases
- Demonstration of real utility or the “so what”
- Finding a sustainable business model
I ended up at Foundation Medicine largely because nearly every mentor I had at the time told me that cancer genomics was where computational science was becoming real in medicine, and Foundation Medicine was leading that charge.
For a lot of reasons, that point in time represented a “perfect storm” for data science to finally have a real impact in clinical care. For one, genomics is inherently a high-dimensional, complex space that was just hitting its stride in terms of broader belief that it was going to have a huge impact on the practice of oncology. There was a sense that genomics would be critical not just for clinical research, R&D and drug development, but in the day-to-day practice of clinical oncology. Additionally, as became evident over the next several years, it was becoming very hard as a practicing clinician, seeing 40 patients a day, to keep up with the incredibly rapid evolution of the field. There were new biomarkers coming out every month, or so it felt, with new drugs associated with increasingly complex biomarkers. It felt to me like it was the first place in medicine that I had experienced where doctors were actually asking for help. In almost every other part of clinical medicine, it was technologists or scientists that were pushing new technology, science and computation into the clinic, meeting resistance from clinicians who felt it was unnecessary complexity in their workflow. But here it was a cry for help to stay on top of the quickly moving body of research. Combined with a high dimensional complex space that was evolving quickly and a fair amount of desire from the clinical community in the clinical applications space, it felt the conditions were right.
Barry: Can we just double click on that for one second, this is fascinating. As an outside observer, it looks like there was a very rapid progression in oncology where you had a long history of academic research suggesting that there were biomarkers relevant to clinical care. But then there was this tipping point where there were large data sets from Foundation Medicine and others that became available, new biomarkers, and clinically relevant therapies that were targeted at those biomarkers; and then exactly what you just described, clinicians that were all of a sudden caught in this flash flood of new information. Is there something unique about oncology? Can we expect that this same series of events is likely in other therapeutic areas too when they hit a tipping point?
Gaurav: Barry you’ve brought up a number of great questions that are so important. Causally and biologically, cancer is a disease of the genome. If you think about contributing factors to onset and acceleration of disease, genomics probably accounts for a disproportionate fraction of that, making genomics really important for clinical decision making. From a drug development standpoint, you have targeted therapies that are tailored to a particular mutation or class of mutations. This is not true in a lot of other TAs where you cannot isolate causal factors to something so computationally ascertainable, so I don’t think it was an accident that oncology was a beachhead. There were a lot of the target therapies that felt like pretty big improvements from a tox profile compared with previous treatments, and some exceptional treatments like the NTRK inhibitors which were truly breakthrough drugs from an efficacy standpoint. Feedback loops were also pretty fast; to run these trials unfortunately doesn’t take very long, because if a drug isn’t working, you unfortunately see progression pretty soon. There’s a lot of things, Barry, that made oncology a particularly optimal place for this all to begin – the primordial soup of data for precision medicine, if you will. So to address your next question, where else in medicine can this or does this exist?
Barry: Yea, what’s the “cambrian explosion” where we expect this to happen in other therapeutic areas?
Gaurav: Yeah, I think that question came up a lot in the wake of Flatiron Health and Foundation Medicine, which were two of the pioneers and anchors of this industry. The question is “Does real world data matter as much in other fields?”
The thing that is interesting about oncology is that the standard of care changes really fast, so real world data companies worry about whether their data set from five years ago will be relevant. In fields that have mostly been static for decades, of which there are some because there unfortunately haven’t been advances in treatment, you could argue that you might not need newer or more current data. But in Oncology where the standard of care changes every 3-6 months, recency is essential. Without licensing recent RWD, you could probably wait for 12-18 months for the insights you’re looking for to be published in the traditional academic research process, but as you know, an 18 month head start is a huge advantage for pharma.
Another area ripe for change, from a precision medicine / genomics standpoint, are fields that are incompletely phenotyped today like neuropsychiatric conditions where the DSM doesn’t really accurately stratify and segment subcategories of different pathologies. As we know, there haven’t been a ton of successful drug development efforts in Alzheimer’s Disease and dementia more broadly. There’s reasonable, if still unproven, reason to believe that you could characterize these individuals better if you took a more deep granular look at their omics data.
Barry: So is the signal that some therapies are wildly effective in a portion of a population but largely not effective in another portion where we have a stratification issue?
Gaurav: Yes, heterogeneity of treatment effects is probably a necessary precursor to all of this. But the one thing that didn’t initially occur to me is that you often see subgroup analysis in an all-comers population where you see overlapping control and intervention arms and some post-hoc analysis shows an area where a subgroup would have benefitted. But if at baseline your treatment and control curves are basically equal, if you think you found a responder population, it necessarily means that the rest of them do worse on treatment than on control, which is hard to justify biologically in a lot of cases. So you probably actually need both heterogeneity of treatment effect and some nominal (if limited) signal of benefit in the all comers population. If you don’t have that, you have to bet on the drug being better for some subset and worse for some subset and the biologically plausibility of this is trickier depending on what the control is. If the control is an effective treatment then maybe this is possible.
I also believe that extending this precision medicine playbook outside of oncology disease areas might not look the same everywhere. In oncology you have this incredibly strong causal pathway between genomics and disease such that if you modify the effect of the genomic alteration, you can maybe impact the disease in a meaningful way. If you look at other diseases, especially as chronic conditions, they are complex both genetically (e.g., with often polygenic contributions), and also have complex interplays between genetics and environmental contributions. The causal impact of the genomics themselves is usually much smaller than it is in oncology for example. That has ripple effects throughout the entire field on diagnostics, drug development and clinical applications and decision making.
In the last year or two, I’ve started to appreciate that there are ways to advance precision medicine out of Oncology that don’t necessarily require the same close causal relationship. Polygenic risk scores (PRS) in cardiovascular disease, is one example of this. PRS have been shown to predict a reasonable, but not complete, amount of risk of developing coronary artery disease in the future. It turns out that if you develop a polygenic risk score mapped to predict future risk of CAD, and you take the top quintile (top 20%) of that risk distribution, you end up with a higher event rate in that population, as you would expect. This means if you are trying to run a cardiovascular trial – which is a massively expensive endeavor with huge outcome studies requiring many years of follow up to see enough events – you can drop the cost by 2x or 3x by enrolling the highest risk individuals, because you’ve increased the event rate which is a linear relationship. More surprisingly (and more importantly), it also turns out that in the top quintile of risk, you see better drug effect (i.e., a greater relative risk reduction) from lipid lowering treatments, for example, and this has been demonstrated multiple times in post hoc analyses.
So what does that mean? It means that the drug is actually more effective on a relative basis as well as an absolute basis and those things compound. You could potentially end up reducing the cost of the development by 10, 20 maybe even 30x. Now you see massive implications for drug development even without having to have the tight causal relationship that you see in Oncology. If you follow this through, you can imagine that biomarkers become part of cardiovascular drug development and clinical care, but for a different reason and mechanism than in oncology.
Barry: So that’s a patient stratification effect rather than a causal effect
Gaurav: Correct, but there has to be some degree of causality since you are getting a larger relative risk reduction which some people might call mechanism attributable risk. But it’s not necessarily the case that the genetics are linked causally to the mechanism of the drug, – that’s the big difference vs oncology. I do think we will see genomics play out outside of oncology, but I don’t think the model can’t be copied 1:1 since there are fundamental structural differences in the mechanism of action of disease. As we figure out creative ways to use the same methods, researchers will probably borrow a lot of things that we learned from oncology. It’s not like we will be starting from square one, we will be standing on the shoulders of the giants in oncology that pioneered these ideas.
Barry: Thanks again Gaurav for this enlightening conversation! Really appreciate you joining me today and for your partnership and advisory support as we drive genomic data forward for research.
Gaurav: Pleasure to be here with you!
Gaurav Singal, MD, is a physician, computer scientist, and entrepreneur who has spent the last 15 years designing and building innovative software and data products that enhance patient care and enable the development of new medicines. Most recently, he served as Chief Data Officer of Foundation Medicine, where he oversaw the design and development of its real world data platform, linking clinical data, digital pathology images, and bio-banked samples with the largest cancer genomics dataset in the world. With this platform, Dr. Singal and his team built a number of industry-leading products including RWD/RWE solutions to increase efficiency of biopharma R&D, novel clinical trial designs to enable broader patient access to potentially lifesaving medicines, clinical decision support tools for complex clinical cases, and a discovery engine to accelerate new diagnostic development.
Dr. Singal earned his B.S., summa cum laude, in computer science with a focus on AI and machine learning from Columbia University, earned an M.D. with honors from the Harvard-MIT division of Health Sciences and Technology, and completed residency training in Internal Medicine at Massachusetts General Hospital. He has authored over 20 peer-reviewed publications in journals such as JAMA, Health Affairs, and Cancer Discovery, and has been invited to speak at dozens of national and international conferences including HIMSS, Stanford Big Data, Personalized Medicine World Conference, and the National Academy of Sciences.
Dr. Singal is currently advising, investing, and incubating a portfolio of companies spanning diagnostics, therapeutics, data science, and venture capital / private equity on topics including computational diagnostics, real world data, and tech-enabled clinical development. He continues to see patients as a physician at Brigham and Women’s Hospital and is on the faculty and Board of Advisors at Harvard Medical School, where he directs a course on product management and health technology innovation.