Listen on
Overview
The healthcare system often acts reactively, diagnosing diseases after significant progression has occurred, leading to costly treatments and missed opportunities for early intervention. This approach leaves executives and data leaders grappling with the challenge of moving beyond symptomatic care to a truly preemptive model that can identify and halt disease pathways before they cause irreversible damage. In this episode, host Ross Katz speaks with Scott Lipnick, Co-Founder and President at Etiome—a venture launched by Flagship Pioneering—who shares how his background in physics and computational biology informs his work in redefining medicine.
Scott introduces Etiome’s Temporal Biodynamics platform, an AI-powered system designed to map disease progression with subclinical resolution. He details how Etiome leverages multimodal datasets, from electronic health records to single-cell ‘omics, to pinpoint early biological changes. The conversation illustrates how this allows for the development of highly specific, stage-appropriate treatments, using the example of liver disease. We also explore Flagship Pioneering’s unique venture creation process and the broader implications of preemptive medicine for healthcare economics and drug development.
Key Takeaways
Subclinical Disease Forecasting Achieves New Accuracy by Leveraging Temporal Data
Traditional diagnostic models frequently fail to detect diseases in their nascent stages. Etiome’s platform uses population-scale clinical data and high-resolution biological samples to achieve 3-4x better performance in predicting conditions like liver steatosis and fibrosis compared to established markers such as Fib-4, AST, ALT, and BMI. This precision enables the identification of individuals transitioning from health to disease with unprecedented subclinical resolution, allowing for much earlier intervention.
Disease Progression Exists as a Continuous Spectrum, Not Discrete Stages
Conventional medical analysis often simplifies disease into binary states or broad categories for statistical modeling. Etiome’s Temporal Biodynamics platform redefines disease as a continuous biological journey, constructing a detailed temporal timeline for each patient using multimodal data, including EHRs and ‘omics. This approach maps cascading cellular and molecular events across progression, revealing distinct therapeutic opportunities at each specific stage and challenging the notion of fixed disease categories.
Reliable AI Models Demand Ground Truth Data to Mitigate Technical Bias
Training AI on raw clinical records can inadvertently cause models to prioritize technical noise or systemic biases, rather than true disease signals. Etiome addresses this by prioritizing high-resolution ‘ground truth’ data—such as labs, medical images, and biopsies—and employing advanced modeling techniques. This strategy differentiates genuine biological progression from irrelevant covariates, leading to more reliable and equitable clinical predictions and ensuring the integrity of medical insights.
Temporal Insights Redefine Therapeutic Windows and De-risk Drug Development
Many drug development efforts falter due to uncertainty about the optimal timing for therapeutic intervention. By precisely identifying the cellular and molecular changes occurring at specific disease stages, Etiome’s platform defines accurate therapeutic windows. This capability allows for the selection of patient populations most likely to respond to a given treatment, promising to reduce drug development risk, shorten clinical trial timelines, and broaden the effective application of existing therapies.
Related: CorrDyn helps biotech and life sciences clients build sophisticated data capabilities. We deliver rigorous AI strategy, tackle complex data engineering challenges, and ensure high data quality for critical medical insights.
Full Transcript
Jason: Welcome to Data in Biotech, a podcast from CorrDyn where we explore how companies leverage data to drive innovation in life sciences. Every two weeks, we sit down with an expert from the world of biotechnology to understand how they’re using data science to solve technical challenges, streamline operations, and further innovation in their business. Today we’re joined by Scott Lipnick, co-founder and president at Etiome, a venture launched by Flagship Pioneering to redefine the way that we forecast, diagnose, and treat disease. Scott and his team are building AI-powered tools that identify subclinical disease progression and enable earlier, more personalized interventions. In this conversation with Ross, we unpack the concept of temporal biodynamics, exploring how Etiome is transforming chronic disease care with stage-specific treatments, and examine what it really takes to translate complex data sets into actionable medical insights. Here we go.
Ross Katz: Scott Lipnick, welcome to the Data in Biotech podcast.
Scott Lipnick: Thanks, Ross. Pleasure to be here. Thanks for having me.
Ross Katz: Awesome. Well, just to kick us off, would you mind giving us an introduction to your background and what brought you here?
Scott Lipnick: Sure. Yeah, so my background is originally math and physics, a little econ, and originally was studying space, then moved into medical imaging, eventually realized that was probably not the space for my long-term career. Had a stint or fellowship at the National Institutes of Health, eventually made my way up to Harvard, and on the junior faculty track there and MGH, spent a lot of time thinking about how to move from physics into more health-based focus and biology-focused, intersected with genomics and computational biology, data science. Eventually left there to go to a restart of a company called PatientsLikeMe after they had been acquired by UnitedHealth Group. At the time they’d been doing this really cool project called Digital Me, where they were going out to people’s houses, collecting blood, collecting a ton of extra non-healthcare-based data, patient-reported outcomes, blood, genomics, transcriptomics, everything, just to figure out how to better subtype patients, how do we better understand what are biomarkers that will be reflective of stages of disease, medication response. It was really a cool program. United bought them to rebuild it and jumpstart it, and I was hired to help lead that initiative. Was awesome, but United is gigantic. We ended up being merged into their venture arm and then pivot a little more into behavioral medicine. At the time I was trying to figure out how do I get back into more quantitative sciences. I’d already jumped from physics into biology and been bouncing around a lot and talked to some people at Flagship who had started an initiative that at the time was called Health Security and has now since evolved into a Preemptive Health and Medicine Initiative. And they were looking for somebody to help them build new companies that would change the way that we find and treat people earlier in disease. That’s how I got to Flagship, and it’s really quite a great four-plus years so far.
Ross Katz: Awesome. Well, it’s great to have you here and since you mentioned the preemptive health and medicine initiative, would you mind just introducing us to Flagship’s preemptive health and medicine initiative and how it’s manifesting as Etiome, the company that you’re leading and that we’ll be talking about here today?
Scott Lipnick: Flagship, if you’re not familiar, it’s a biotech company. Has history in venture creation. We basically have a group that accumulates resources and then as scientists internally we ask our leadership for funding to start companies. A lot of the whole process internally is thinking about ways that we can make major impact on health and medicine. The Preemptive Health and Medicine Initiative largely came out of early stages of the pandemic thinking, infectious diseases are taking over. How do we make sure that our society is better protected from them? But if we extrapolate beyond that, how do we also think about these slow-burning pandemics like chronic diseases and cancers and what can we do to make sure that we’re better protected in advance of these diseases? That was the Preemptive Health and Medicine Initiative. If you go to our website, you can see lots of essays on it, lots of things that happened before my time at Flagship, but ultimately how do we convert all these ideas into tools and products that can actually be impactful in people’s health? Etiome is the first company that was incubated within that initiative. There are a few other companies that are part of it that have since been created or were created before that really already fit the mission.
Ross Katz: Awesome. Well, so I think we mentioned Etiome as embodying a preemptive model of medicine. Can you just unpack what preemptive medicine means in practice and maybe highlight some of the ways that data or data science intersect with that preemptive approach?
Scott Lipnick: Yeah, so reminder, background in physics. These are data people that we’re talking to. I’m going to try and define some words, and the definitions are probably not accurate, so just go with my intent as opposed to the actual meanings. But we were really thinking about preventive medicine versus preemptive medicine. Preventive medicine, we were thinking, is primary prevention. How do we stop people before the disease is ever around? Really a protective shield where preemptive medicine, we were thinking more about the delta between biological changes and symptomatic changes. How can we find diseases that have already started? We can already quantitatively define that they are a disease or they are some stage of dysfunction and treat them before they’ve caused irreversible damage or caused symptomatic presentation. That’s how we think about preemptive medicine: what is the space in which we can find definable changes in your body that we can alter and stop or reverse progression of disease? For Etiome, the way it happens, we have these explorations at Flagship. You go through all these what-if questions, a thousand what-if questions. One of our what-if questions was, what if we could take the sickest person on earth and actually bring them back into health? And the flip side of that was, what if we could preempt progression of disease at any specific time? We kept on toying with these ideas of, all right, if you’re extremely sick with a complex disease, you probably need to have lots of different medicines that are going to slowly pull you back into health. That started to lead into the idea of, all right, how do we pair that idea with if you’re very early in disease, how do we find the exact medicine that’s going to be right for you? We just kept on exploring what is the technologies that allow us to figure out which medicines would be right for which stage of disease, which combinations of medicines would actually be able to push you back as far as possible? That was the seeding ideas behind Etiome.
Ross Katz: Awesome. And in practice, what does that mean for Etiome’s approach to preemptive medicine?
Scott Lipnick: We figured we needed to do three main things. The first is, we needed to build some pretty cool AI tools. Well, tools, turns out AI is the right way to go about it, but we need to figure out how to identify who is at risk or likely progressing from health through different stages of disease. Right now, the healthcare system is largely based off of people nominating themselves, or in rare cases, broad-based screening, to identify people that are transitioning. We thought, we have decades and decades of electronic health records, we have a lot of information on people who were diagnosed or not diagnosed and that temporal information gave us the idea that we could forecast and identify where people are at in progression at a much higher subclinical resolution than possible. So the first pillar is, can we identify those people that are transitioning effectively? Now, the second is, can we confirm that their biology is changing? We put a lot of effort into understanding which biomarkers will be relevant to confirm that the cells that are actually being affected and the organs are actually dysfunctioning in the way that we want. So the first step is identify people who are likely transitioning, confirm that their biology in the state that they want, and then finally, developing stage-specific medicines that will treat people exactly where they are. In practice, we work backwards a bit, we start with the tissue and we say, what’s happening here as you progress from health through disease? We use an approach that we say is like the Framingham Heart Study. That was an incredible longitudinal study that showed how do we monitor health through different stages of disease. Now, it took forever and in today’s dollars, I don’t know, billions and billions and billions of dollars, but it manifested in probably one of the biggest drugs or biggest impacts in the world in statins and identification of cholesterol. We thought with these technologies that we have today, is it possible to do that instantaneously? With effective subclinical labels, with access to population-scale tissue and blood, we’ve been able to figure out at the single-cell level, at the tissue level, what are the cascading biological events that are leading from health through different stages of disease, ultimately through end-stage disease? How do we figure out which of those are going to be presenting themselves as potential biomarkers, really circulating proteins is what we’re focusing on? And then which of them can we use our causal inference models to identify are therapeutically relevant and, maybe a little more futuristic, how do we think about combinations of them that can actually bring people from any specific stage back into a fully functioning regenerated organ or person?
Ross Katz: So, everything that you just described, I’m assuming that’s the elevator pitch for the temporal biodynamics platform. Am I understanding that correctly, or are there other elements of the platform that you would bring into the conversation?
Scott Lipnick: That’s a good question. I think those would be the three core pillars. There’s other elements and depending on how deep you want to get into anything, I’m happy to open up, but those would be the primary three.
Ross Katz: Yeah, so you’ve got this platform that’s reimagining how we look at medicine in terms of identifying what stage a given person is with regard to the progression of a given disease. And it sounds like because it’s a platform, to a certain extent, this methodology can be made somewhat disease agnostic. So I’m just curious about how the idea of a platform that approaches disease analysis from this perspective, from this temporal perspective, came about and to what extent that platform can be made disease agnostic to accomplish that kind of goal?
Scott Lipnick: I think if you look back at the majority of studies that have happened over the history of science, a lot of it has been juxtaposing one or two stages versus one or two others and really thinking about discretizing health so that we can look at biology in a way that statistics works. It’s a lot about labels, it’s a lot about scale. I think over the past few decades, we’ve been able to, or not we, the field or many of the fields, whether it’s advertising or science or others, have been able to identify probabilistic models that can very specifically isolate signals from extremely complex data sets. We started to take the learnings from that and said, instead of thinking about health and disease as binary states, what do we need to do to transition our analyses to actually focus on complex real-world populations? A lot of that is number one, getting access to huge amounts of clinical data. So we have roughly 200 to 300 million electronic health records, really deep insights on people that are from them. However, they still have a ceiling of value. We put a lot of effort into getting high-trust certification, which I’m guessing a lot of people won’t know or care about, but it’s important because it allows us to get behind HIPAA firewalls, or at least HIPAA-protected data, and we’re extremely secure with every data that we would ever touch. But in that data, there are medical images, there are free text in clinical notes, and all of that information gives us the ability to paint a much more effective picture of how people at scale across the population have transitioned from health through stages of disease, ultimately allowing us to build models that can generate subclinical, so really higher-resolution labels for people across a number of different dimensions. With that, we’re now able to take hundreds of patient biopsies or blood samples or thousands of them and label each individual not only across a continuous spectrum of disease, allowing us to order them with the disease we’re interested in, but also allocate signal from many covariates that typically are discretized into tried to be bucketed within healthy and disease. Think comorbidities, genetic propensities, or in worst cases, people have tried to have extremely homogeneous cohorts that really aren’t reflective of the general population, just men or just white people. We’ve built a system that allows us to take in really complex comorbidities, covariates, and figure out how we can remove that signal and ascribe a continuous representation of the biology of interest that we are interested in. It starts with clinical data, population-level data. Then we go into largely transcriptomic and proteomic data from tissues and blood. And we figure out at the single-cell level the order of those cells and the bifurcations of different subpopulations, and we use that to learn the biological insights that help us find biomarkers and therapeutic targets. And then which of them can we use our causal inference models to identify are therapeutically relevant and, maybe a little more futuristic, how do we think about combinations of them that can actually bring people from any specific stage back into a fully functioning regenerated organ or person?
Ross Katz: That makes sense. So you’re taking all of these different data sets, the multi-modal population data that you just mentioned that you’re getting from the EHRs and then the omics data, the blood samples, the proteomics samples, and you’re trying to get this timeline of each patient’s history. And then my understanding is you can look at that through the lens of any particular disease category in order to understand how the timeline that you’ve created relates to that disease category. Am I thinking about that right, or where is that model off?
Scott Lipnick: You’re exactly right. So we are at the cellular level figuring out which in which people are which cells first. So if you model disease at the cellular and molecular level, you’re going to be doing it in an across patient way, not any individual has one type of cell at one time. And so we try and figure out within any individual we might be able to treat. So think in Alzheimer’s disease, you’ve got mild cognitive impairment. These individuals likely have pretty high burden of A-beta around a number of their neurons. They’ve got accumulating tau, but still on the upswing of tau. But really they don’t have much cognitive decline, just a little bit. And so we’re asking at that stage of disease, what do their cells look like? What’s happening in their oligodendrocytes or the astrocytes, their neurons, their microglia? And then as people progress further into cognitive decline, we might see a plateau in A-beta or tau. Some people may progress faster, some people may progress slower. We want to know what are the biological changes that are associated with those different subtypes. And as they progress further into disease, what are the changes that are happening then? And if you think about disease and you think about the exciting targets in Alzheimer’s, we’ve got some in the microglia that are related to inflammation, we’ve got some in neurons related to hopefully neuronal resilience. We’ve obviously got A-beta that’s proven successful. Our hypothesis is that depending on when you engage a person, the effective therapy is going to be extremely different. Because the cells that they have, the accumulation of them or the aggregation of them are quite different as you progress through disease. And I think that bears out in Alzheimer’s. If you look at the high level of variability of response to medication that people have, we even have seen things like women and men have drastically different cellular responses to A-beta accumulation. If you look in clinical trials and you see similar impact, right now a lot of that information is masked or just uncertain. We accept the fact that there’s large error bars on our clinical studies. We as a company think that there’s a lot of features that we can use to resolve some of that uncertainty. Probably not all of it, but a very good chunk of it, and our initial results are implicating that to be true both in Alzheimer’s and fibrotic liver disease, we’re leaning more into many other diseases, immunology is a space where I think you see huge amounts of variability in patient response to medication and the question should always be, I’ll bring it back to math here. In any sort of a system, there should be ways to be able to ascribe what’s driving variation in that signal. I think we’re just at an early stage of science, where we still don’t know much of it. Etiome is at least taking one step towards that, which is bringing in temporality and patient substaging and subtyping to try and reconcile that. Our hope is that we can show you some higher probabilities of success of the targets and drugs that we’re developing and companion biomarkers that will help us really confirm those populations in the short near-term future.
Ross Katz: No, that makes a lot of sense. And so what I’d like to do is using either Alzheimer’s or another disease category of your choosing, you mentioned three steps of the process: identifying where the transitions are, what are the transitions that occur, confirming that the biology is changing with the biomarkers that you’re identifying, and then developing stage-specific medicines. So if you’re willing to walk us through the intuition or the thought process that goes into each of those for a given disease category, I think that would help to elucidate how the platform works.
Scott Lipnick: Okay, maybe I don’t want to cause any whiplash, but why don’t we switch over to liver for a second because I think it’s an easier disease in terms of clear stages of progression. It’s also one in which we’ve made a lot of progress, particularly because of the accessibility of the tissue, and the ground truth that exists for staging. In MASLD, a disease formerly known as NAFLD and NASH and MASH, basically patients accumulate fat in your liver. It’s called steatosis, as they accumulate fat there’s a semi-nonlinear relationship between the onset of inflammation, the onset of fibrosis and ultimately the three of those phenotypes end up building and building until your liver can’t handle it anymore and you end up having a cirrhotic liver, really a dying liver, and that progresses into hepatocellular carcinoma or liver failure. So pretty dramatic disease, hundreds of millions of people or about 100 million people in the US are affected by some sort of MASLD or MASH, millions and millions of those people have fibrotic livers and a good number of them actually convert to liver failure and cirrhosis and obviously a subset of that into cancer. So what we set out to do was say, number one, if you look at electronic health records, it’s a disease that should be 30 or 40 percent of the population, there’s about two to four percent of the population that have a diagnosis. So it’s really an invisible disease, partially because there haven’t been any drugs until this year to treat it, at any stage of the disease. For the first part we asked ourselves, how quickly does this disease progress and which people who don’t know that they have the disease can we identify that have it? We said, all right let’s partner with a clinical site, we actually worked with Vanderbilt University on this study. We built models across all of our data that said who is going to have different stages of steatosis and fibrosis in particular and can we go out and identify people who have never been in the clinic, never seen a hepatologist, and accurately predict the exact level of steatosis and fibrosis that they have? We are preparing a publication on this work, it should be out pretty soon. I won’t go into too many details other than to say about 3X or three or four X the best-in-class models that use FIB-4, AST, ALT, BMI. We’re able to do much better at identifying at a high resolution where people are in disease progression, really confirming that we can identify and forecast who is where in progression. We were able to accumulate hundreds of liver samples that represented the entirety of disease progression, we single-nuclei sequencing on them to get single-cell transcriptomic resolution. We used the subclinical, really the percent steatosis, percent fibrosis labels to order those cells across people. The key element is that different people have different windows of progression and so we can map them together similar to shotgun sequencing and really reconstruct the biology that’s specific to the steatosis and fibrosis across different dimensions, really deconstructing all of the other comorbidities. Using that data, we were able to nominate panels of biomarkers, some that are very specific to how the disease progress in different cell types, some that are really just focused on hepatocytes or stellate cells and used as calibration markers, and we used those to confirm that we can really selectively use biomarkers to identify where people’s cells are across disease progression. And then we’ve nominated a number of targets that are specific to different stages of progression, both in hepatocyte and stellate cells and macrophages, or sorry, Kupffer cells, and shown that they can reverse disease in vitro and more simple model systems in vitro as well as in organoids as well as in some of our proprietary assays that are mimicking different stages of progression as well as gone into preclinical mouse models and shown that at different stages of progression, we can selectively reverse progression given different targets that we’re addressing. Here, in totality, that means that we can go out into the wild, or the real world, sorry, whatever you want to call it, and identify people that are at different stages of disease, we can selectively isolate synchronized populations that are at specific stages, we can do that both with clinical data as well as confirm that with biomarkers, we can now selectively give them single or combination medicines to help bring them back as far into health as possible. We have the actual medicines that we can use to do that. We’ve novel targets that we’ve identified, we also have novel targets that have really interesting repurposable chemistries that we can put together, reformulate to actually bring into the clinic. And now we’re at the stage where we are talking with different pharma companies and different partners about how do we take some of these innovations and move them forward to actually help patients today and seed a better pipeline of innovations for the future.
Ross Katz: In your conversations with different partners about taking these innovations and helping patients today, what do you view as the way that Etiome’s temporal biodynamics platform is going to be able to bridge gaps that weren’t able to be bridged before, or how do you envision those partnerships working?
Scott Lipnick: That’s a great question. There’s a lot of different types of partnerships and actually for your audience, we launched the company about within a month of recording this podcast. So we’ve only been publicly visible for a while. We’ve obviously been having conversations with different key partners for longer than that, and the company’s existed for about four years. We have a lot of different types of partners. We have partnerships with clinical centers where we’re thinking about how do we create a new pipeline of research, new sources of revenue from clinical trials, just rethinking about how patient engagement and enriched nomination can work. We have partnerships with or we’re exploring partnerships with clinical trial developers. I can’t name any names now because I don’t know what I’m allowed to, but there’s a lot of great groups out there that run a lot of clinical trials that we’re thinking, how do we bring our innovations into them, both to be thinking about enriching patient populations as well as bringing in new biomarkers that can be used to synchronize populations as well as imagine how we can get them through the FDA as metrics of efficacy. Really something that could make trials much shorter, much more effective, much more biological. And then with pharma partners in particular, we’re thinking, how do we identify the most relevant target or targets for their specific interest, the stage of disease that they’re interested in? In fatty liver disease, we’ve got groups that are talking to us about F1, F2 fibrosis, we’ve got others that are specifically interested in F3 and the F3, F4 and the conversion into cirrhosis. And what’s cool for us is each of those represents very different target profiles. Where you once had a focus on disease, we’re now bifurcating that into many different sub-diseases, taking these diseases and converting them into syndromes that have different profiles. We’ve seen a lot of strong interest from pharma about how do we take the insights we have, the liver in Alzheimer’s disease, and completely even shift into different disease areas and say, what if we were able to treat Parkinson’s this way? What if we were able to treat rheumatoid arthritis in this way? What about even thinking of cancer? Yeah, I know I’m going on for a long time, but I’m really excited about how the technological advances we’ve made and the team we’ve built, which is awesome, by the way, people are great, how do we pair with the large institutions that have incredible institutional knowledge that we just don’t have and will never have at the scale that they do across therapeutic areas, across clinical partners. So yeah, I think the conversation’s going well so far.
Ross Katz: Right, so it strikes me that, just taking your pharma partnerships as an example, you might in the non-temporal approach to this, you might develop sort of a single, you’re looking for a single drug to target a single disease, and then you’re coming up with the regime for treatment with that single drug for people who are in all of these various stages of the disease, but only post-diagnosis, if I’m thinking about that right. Whereas when you have these biomarkers, there’s this opportunity to both bring treatments to bear that wouldn’t have been there otherwise because you’re preparing to treat people before they’re actually diagnosed with the disease, but then also there’s this opportunity to think differently about the disease at these different temporal stages such that potentially there’s more than one therapeutic approach that’s being brought to bear at those different stages. Am I thinking about that right, or understanding you correctly?
Scott Lipnick: I think so. But let me try and make sure that we’re both thinking about the same way. Sure. Long-term vision, we want to treat people as early as possible. And I think every company, every group we talk to also is interested in early intervention. I think there’s also a lot of groups today that have uncertainty around the therapeutic window of relevance for their targets. If you think about genetic studies, you’re basically saying this gene is likely relevant in this population in this disease. I know that it’s been different their whole lives. They have this predisposition. At some point in their lives, the product of this gene is making them either more protective or more susceptible to a disease. And the question is always when is it most relevant, when is it therapeutically active? If you look at a lot of the targets that come out of GWAS and genetic studies, they are better than most, but they also fail quite a bit. I think that a lot of that failure, and that people have been bringing to us, is because they just have uncertainty on when is the right time to intervene. So where we’ve been seeing some traction is saying, for any program, how do we give people more information about the population that’s likely to respond? That can be from taking a drug program that’s preclinical trials and saying let’s help you design the actual target population. It can be as simple as just different stages of disease. It can be focused on specific biomarkers. As well as, for programs that are working now, can we even think about how do we expand the population that you’re treating because we know that this is actually active earlier? Or how do we really do something even bigger and think about you just had this major drug that’s out there and you’re going to go out and try it in every other possible inflammatory or fibrotic disease, can we be a little bit more intelligent about that approach for indication expansion and say, all right something worked in kidney fibrosis, now when do we look at liver fibrosis or lung fibrosis? Well, it’s not going to be the exact same thing, so we should spell out by doing the science and figuring out which populations are actually likely going through the processes that would be relevant for your intervention. Our vision long-term is to be able to go as early as possible. Our broader vision is to take anybody and figure out what are the cascading events that need to be treated in sequential order or in combination to be able to get people back to complete health. But to get there, we’re actually just building tools that anyone can use that help understand when a target is most relevant and how to achieve the highest probability of success of their program once you get into actual people.
Ross Katz: Yeah, that makes a lot of sense and I want to return to the example you gave around the lead up to liver failure. You mentioned that two to four percent of the population has this diagnosis for MASH or MASLD, if I’m remembering correctly, but maybe 30 to 40 percent of the population might be indicated for it. And then you mentioned that by understanding how the disease progresses and applying the temporal biodynamics platform to it, you were able to get substantially greater performance relative to the best-in-class models that are out there. I’m curious from your perspective, to what do you attribute your ability to get best-in-class performance? Is it this temporal lens? Is it the quality of the data that you’re bringing to bear and the care that you’ve exercised with that data to align it with that temporal approach? Is it the emergence of these causal methods that maybe aren’t being used as broadly as they might be for this kind of exercise, or is it something else that I’m not even thinking about?
Scott Lipnick: We take different approaches when we’re trying to build models for a specific person versus trying to build models to accumulate a cohort, and so one of the things that we did in thinking about therapeutic studies, clinical trials or biomarker validation studies, was let’s think about the target population we need at the end, the overall distribution of them. And if we build probabilistic models for each individual, now we don’t need to be 100 percent right on each person, but if we get enough people then in aggregate the presumed distribution is going to be what we want. I would say the team has done a lot of really interesting work in building new models that help us rethink the way that we’d want to approach engagement. Second, I’d say we take a really focused approach on how do we get ground truth data for any indication we want? So if we’re studying MASLD and MASH, we’re not going to go into clinical records where there’s only two to four percent of the population that have it and say who got it. That’s actually going to be looking at a marker of who had super severe disease and was in the hospital for something else and ended up getting a diagnosis, who had some sort of cancer and this was just a secondary marker of it. In particular, it also bleeds into some health equity issues because people who have data tend to be just the better-off, the more health-literate people in the population who have resources. What we did was we said, all right let’s find as much ground truth data as possible. That would be in the form of labs, in the form of images, and in the form of biopsies. And say, how do we figure out what is the difference between the population that we have ground truth on and the real-world population and figure out a transformation of the models we built on those high-resolution cohorts and make them applicable to the real world? And I’d say there’s a couple more components. One is, we know that not everyone’s going to be modelable. There’s a lot of people who are just too data sparse. And so we’re putting a lot of effort in looking forward into what are user interfaces that can be developed that would help collect very light low-touch information from people that make them more eligible for our algorithms and our forecasting. So I would say better data or really high resolution data, thinking about the way that we model probability distributions is quite novel. Thinking about how we take variation in the distributions of the populations from whom different data sources come, and then ultimately trying to build capabilities to get more data from people outside of the clinic.
Ross Katz: That makes a lot of sense, and one of the things that jumped out at me was that you said that you’re almost masking from yourself, you’re using the richness of the EHR data but not just naively training a supervised model on the EHR data that you have, but going from first principles about the disease and what you would expect to see and leveraging the data at your perspective to construct that model of causality from the data that you have, and then using that to guide, I’m assuming, what other data that you would like to gather from the population in order to better understand and fill out that model.
Scott Lipnick: I think as data people, I think everybody nowadays is becoming a data person because there’s so much of it and there’s richness in it. Our first task should always be: what are the technical sources of signal that are in here? What are the things that could be not of interest to what we’re doing actually for the biological signal? I always tell them, our job is really to figure out what is the signal that you don’t want to know about. And eventually if we can get rid of most of that, then our goal is to give you exactly the biology that you want, but really say it’s 80 percent, 90 percent enriched over what would have otherwise been complex data. In electronic health records, it is not equal. A lot of times if you go to get a colonoscopy, the doctor is just going to click colon cancer. When I was looking at this is going to go off on a weird tangent, but anyways I did a study on carriers of different diseases. One of the hypotheses we always had when I was at Harvard was that carriers of recessive disease genetic diseases are likely protected from something. I think I’m not going to go through all the different history about it, but you should read about it, but most evolutionary bottlenecks have resulted in people who have some protection from something and a lot of times if you end up being homozygous for the mutations that lead to protection, you end up having a horrible disease. I’m Jewish, and so there’s a lot of diseases, Tay-Sachs, etc. in our populations. Anyways, the hypothesis was that there is genetic variation that could be mined for protective factors or could lead to different insights. When we’re actually looking in the general population, we want to try and figure out what is relevant to the disease or relevant to some genetic variation in society or what’s relevant to the fact that half of these people were taking a medication and half of them were not? Or what about the fact that some people live in Omaha or a suburb 50 or 60 miles away from there or 4 hours away from there versus me who is living in a suburb of Boston? We have very different access to care. Also in some areas, you have primary care and secondary and tertiary care all together in one site, so your EHRs are unified. In Boston, if I look at MGH and Brigham, a very large percent of the patients that I used to study only came for specialty care, so we missed everything upstream. Anyways, the point that I just belabored is that you have to be very smart, or at least very intelligent about thinking about the patients from whom the data came, thinking about the doctors and the amount of time that they’re spending on each patient that’s generating that data. I’ve been so happy with our team that push a lot of this on me and say, we can’t do good modeling with one data set. We need to be very focused and understand what are the realities of each, how do we get more information about environmental exposure, how do we get more information about health literacy of the population. As you can imagine there’s lots of different variables and in order to be able to do effective modeling, we need a lot of data from a lot of divergent sources and more importantly to understand what are the differences between them and then when we go to a new site, how do we reconcile all those differences to make sure we have the most effective model that’s personalized to each person.
Ross Katz: Yeah, no I really enjoyed that tangent and it brings up one of the things that always comes up for me which is that the better you understand the generation process of the data set you’re using to make a model, the better you are able to account for the assumptions of that data generation in the model that you’re developing using that data. It sounds like deep understanding of how the data gets generated and what might be missing from that data set that might help to paint in the edges of the picture that the model is trying to paint, is an important part of the modeling process as well.
Scott Lipnick: Yeah, and this is I don’t know another tangent then I’ll be done. When I was in academia still, we did a study. It was through our group and a group at Google Accelerated Sciences that was trying to figure out how can we use AI to understand or identify phenotypes in patient cells. IPS-derived neurons as well as fibroblasts from people with a very severe disease, spinal muscular atrophy, versus those that are healthy as well as basically the same people who had their mutations genetically modified to become controls. We found all these really cool signals, we had dozens and dozens of cell lines, and actually this is I think some of the coolest work I ever did. But at the end of the day, the conclusion was the strongest signals were where in a plate you were, which well in the plate you were, what the source of the cells were, the controls had come from one site, some of the severe patients had come from Children’s Hospital in Boston, some had come from Children’s Hospital in or Columbia. And it was just all of these things that are irrelevant to the genetic mutation in the cell, but just of how they were accumulated the first day, how long between biopsy and storing, and just the amount of data, the way that AI could pick up on the specific signals was incredible. But it really just showed that we’re not quite there with being able to control for all of the sources of signal we want. Once we did control for those, we were able to start to get more signal that was related to disease, but it was just such a small blip, the disease-associated signal versus all of the technical factors. And what’s crazy is I thought this should be like Nature or Cell or really awesome and everyone’s like, but you didn’t find anything specific to disease. And I’m like, what do you mean? We found everything related to how we’re supposed to be studying disease and how you’re supposed to be doing science and it was great work. So that’s what I try and tell the team is that our job is not to be the one to hang our hats on finding the best disease signal to the best anything related to disease. It’s really to prove that no one’s going to come in and find some source of signal that is technical or that we just didn’t think of from a comorbidity, covariate, whatever sense.
Ross Katz: I love that and I could keep talking about this for a very long time but I want to be sensitive to your time. So I would love to transition into how Flagship Pioneering does venture creation in your experience and then how that worked in the case of Etiome, because I think that Flagship’s approach to this is really interesting. I know that many listeners are probably familiar with Moderna as this flagship success story and also came out of the venture creation process but would just love your perspective on the history of that process for Etiome and what were some of the milestones that you had to go through in order to get where you are today?
Scott Lipnick: Apologies in advance that I did drink the juice since I’ve been here and I love it. I really do love Flagship. It’s the first place I’ve ever worked where we are encouraged to think really aggressively about ideas that probably won’t work, and to pitch lots and lots of ideas that have medium chances of success but if they work would be transformative. It’s the first place that has, I have bosses that have access to resources that I really couldn’t fathom before and not just financial but intellectually. We have former heads of the FDA, former heads of R&D from major pharma, just brilliant people all around and so I just say it because one, if you’re interested you should reach out, it’s a great place to work, as well as our ecosystem companies. But it’s a place where for the first time in my career, I’ve felt like I don’t have an excuse if we fail, it’s either on science or on my and my team’s ability to manage and operate. And so that’s, if you’ve been in pretty cool places with lots of resources, I would say most of you at least for me haven’t had many opportunities to be able to say that and so I finally be able to say that. Anyways, okay whatever. To the side. At Flagship we have these exploration teams or origination teams as we call them where we think about what are major ways that we can change the world and if successful wouldn’t have any question about their financial viability. For Etiome as I mentioned earlier, we were talking about what are ways that we could treat the sickest people to get them back into being completely healthy? That converted into a story more about how do we treat the healthiest people or the people who are at the precipice of disease in a way that was like statins and cholesterol? Can we turn statins and cholesterol paradigm into something for every disease? The latter ended up winning out as more interesting for a company and also more aligned with our preemptive medicine initiative. But the science was true for both. The way it worked was we had a team, myself, Katherine von Herrmann, Torben Straight Nissen, we were the original crew that was pitching this idea and I’ll say Katherine and Torben had done way more than I did, I joined last minute into their exploration and just took all the credit for it, I’m the one here but they should really be up here. We pitched an idea to Noubar and the rest of the partners and said, I bet there’s enough data out there to prove out that we can subclinically label people, number one. I bet if we get tissues from across disease progression we can rethink the way that people are understanding time and that we could validate that with existing data, some genetic studies in mice or something else that would say that four out of ten of our main hypotheses might already be proven true. And finally, I bet if we look at public data in biomarkers, some of our hypotheses might prove out to be true. So we said give us a couple million bucks and give us nine months, something short, we will be able to prove the foundation of this that we can hire four or five people, have four or five people from Flagship work on this. There were some ups and downs in that first bit of time, but it’s probably the most fun in any company’s timeline that first what we call Protoco, really we went from exploration to Protoco. Once that proved that we couldn’t prove ourselves wrong, and had some good hypotheses then we go back to the partners and we say, I think there’s something here. Give us a magnitude of order more resources and let’s go out and hire a real team, bring this up to twenty some people, let’s start doing a lot more work, let’s start running our own proactive prospective studies, let’s actually think a lot harder transition the programmer from me to really good people, all of the legacy code I wrote is not even stored anywhere, that’s how historic it was at the beginning. Although it was pretty good and I can just tell you the actual algorithms weren’t great, but the annotation was amazing which people should appreciate if you do code at all, annotation’s important. Then we scale it up and then eventually a couple years in then we say we think this is ready to go live. We think it’s ready to start partnering with external groups, we think that some of our initial data is there. That’s when we get a bigger bolus of money, that’s when we start to launch the company, that’s when we start to look for CEOs and CSOs and that’s the stage that we’re at now with Etiome, we have 20 some people in the dry and wet labs, we have a good number of people from Flagship that are working full time on this company as well. Also, I’m the luckiest entrepreneur in the world or founder in the world because I don’t have to deal with finances or legal or HR, I don’t have to deal with hiring any of those people. We have those systems within Flagship that are just there to be fractionated and deployed across all of our orgs and so we just have to focus on science and the only technical thing we do have to deal more is cloud computing because if we don’t own that ourselves then the cost goes crazy.
Ross Katz: So you have this idea, you pitch the idea to Noubar and the rest of the partners at Flagship, and they give you some money to validate some hypotheses. Can you just give us a sense of when you reach that second stage if I understand by moving from exploration into Protoco, what was some of the things that you needed to prove or the data that you needed to gather, the evidence you need to bring to bear for the Flagship team to have confidence in continuing investment in you?
Scott Lipnick: Yeah, okay let me get the lingo right. So exploration is just internal, no money. Okay. Protoco is the first stage and then that transition to Newco is I think what we’re talking about. Got it. This is actually where one of the other co-founders, Avak Kahvejian, started to come into play. He’s a general partner at Flagship, brilliant guy. Started to help thinking about how can we make this a real company, how can we focus on not just proving out some science but proving out or building a platform and a pipeline. We had to show one that there was signal in the data and that we could generate partners that would work with us in this space. We had lined up a partnership with a clinical site to run a prospective study. We had nominated a number of targets that we had validated in vitro that we thought would be transformative in the way that disease is progress and that we thought were specific to different stages of disease. So we had basically plan for mouse studies that we were ready to go. And then on the biomarker side we had shown that some of the initial signals that we saw from the tissues were actually being realized in the blood. It was still pretty noisy at the time but we said here are some hypotheses on how we can clean this up, how we can make it better, how we can develop our own calibration tools. And if we push these things forward, we think that we could treat any stage of MASLD. Now, I’m going to add to that that we’ve also been talking with a number of partners, been thinking about business models, here are the other diseases that we think we should go after in order, we are a business. We have to make money at some point, have to have partnerships. Sometimes I forget that and just want to be doing science but I get reminded. With these plans, if you give us X amount of dollars this is the milestones we’ll hit in the first three months, here’s the milestones we’ll hit in the next six months and so on and so forth and by basically de-risking the science and then showing that we had a clear approach both for scaling the platform, building a pipeline as well as generating IP that’s going to protect all of our foundational intellectual property as well as the targets we’re going after, the biomarkers we’re going after, with that and a lot of rounds of trying to convince and being poked at, we convinced people that it was worth investing in. And to anyone who’s interested in science or entrepreneuring or whatever I’d say one of the things that I never expected was how much positive and negative feedback I would get from people who end up investing in our companies and want to push it forward and it’s like if someone cares about what you’re doing, they’re not going to tell you how great you’re doing. They’re going to tell you exactly all of the things that need to be done differently, need to be done better. And the amount of learning I’ve done since people started investing their money in the stuff that I do, really transitioning from academia into industry and into biotech startups in particular, it’s awesome. But it’s not great for the thin-skinned folk.
Ross Katz: Yeah, that makes a lot of sense. And at the end of all of this, my understanding is that you’ve received fifty million dollars to advance the temporal biodynamics platform at Etiome. So I’m just interested in how are you thinking about deploying that capital over the next period of time, and at the end of that, where would you like to see Etiome be as an organization?
Scott Lipnick: We’ve done a lot of work on platform validation. I think that there’s going to be, I hope, announcements in the short term that relate to other groups partnering with us to help support some of those programs move forward to the clinic. I think fifty million dollars is great and a lot of money but it’s not enough to get a drug into the clinic obviously and it’s not enough to build a pipeline on our own. So we are very focused on getting partnerships across all of our different areas, both because it’ll help resource it as well as because as I mentioned in the beginning, we don’t have the scale to have all of the institutional knowledge or, there’s a lot of information out there and every partner that we’ve been talking with thus far has been transformative in the way that we think about different therapeutic areas and patient populations and the regulatory agencies. We are going to continue to build different therapeutic programs, largely in partnership, and then we’re going to seed some new innovations on the clinical engagement side, on the biomarker validation side, on expansion into different therapeutic areas where we have a little bit more uncertainty with the money that we’ve been able to raise to push for it. I think if you talk to anyone how you use equity investments have changed a lot over the last few years especially the last decade. We’re being very conscious about what we do to validate platform, to build new resources versus scale up into drug development and clinical development and where the line of partnership exists.
Ross Katz: Yeah, that makes a lot of sense. I know that there’s this delineation between the vision for what the company could do in the long-term future if you’re able to scale up and take this methodology that you’re using to its natural conclusions and the very real tactical commercial responsibilities that you have to the organization in this early stage. So I’m just interested in can you color what that vision looks like and then how do you think about that progression from the immediate to the vision that you’re trying to achieve?
Scott Lipnick: I’ll first say within Flagship and the ecosystem there are companies at every stage of disease life or progression, or sorry, company life and progression. So we have learned a lot that as you get to different tiers, the amount of focus on innovation changes a lot. And one of the things that Flagship as a whole has been pushing is to make sure that we as majority owners of most of these companies are pushing those companies to continue to drive the next round of innovation. How we do that and how we go out and get investors and focus on platform companies, asset companies, etc. That’s a little above my pay grade but I’m learning a lot about it. But one of the things that’s critical is making sure that there is a pipeline of new innovation. We’ve talked about it earlier in this podcast, I think that the future is figuring out how to treat people today as well as whatever they were yesterday and the day before, going systemically through the disease journey to actually bring people back to health. I think that there’s innovations that are required to take biomarkers and move them from something that can inform development, that can inform patient selection, to actually being readouts of efficacy. I think those are worthwhile investments for our future and the future of healthcare and the way to differentiate Etiome from other groups. That is where I want to push a lot of our resourcing into is thinking about that future and that long-term vision, while then partnering with people to take the advances we’ve already made and making sure that they are received well and they can turn into an impact, they can lead to higher probability of success and ultimately lower costs of clinical trials and actually treating people before they’ve major debility and changing the way that healthcare is or disease and debility are stifling our economy and anyways I won’t go into the financial ramifications of it. But I think you ask really good question which is how do you think about long-term vision and that’s something that has been ingrained in me since I’ve been here is we can’t just be focused on what innovations we’ve developed for today, we have to have a whole pipeline that follows them up and builds the innovations of the next generation of medicines, both within our companies as well as in the next companies that we’re already ideating.
Ross Katz: Are there any upcoming advances in data, in science, in automation, in policy and technology that are out there that you think of as intersecting with Etiome’s trajectory and potentially enabling you to accomplish your vision?
Scott Lipnick: Well, I would ask that there are some. I don’t know if they’re going to happen. From uncertainty comes a lot of good and bad things. And I’d say right now we have a lot of uncertainty in NIH, in HHS in general, in the FDA and one of the things that I’ve always been hoping is that we can take more creative ways to allow drugs that have longer horizons of trials to move forward. We’ve been talking with regulators and just different sorts of groups. I was at United for a long time. How can we rethink the paying structure of drugs that maybe prove safe, prove short-term or early efficacy, maybe at a biomarker level they could be demonstrated to be efficacious, and then have some sort of alternative phase four structure that would allow long-term recoupment of investment but still more partner with payers and providers to prove that drugs are efficacious and to prove that biomarkers are actually the right way to be studying diseases. And I think we’re building a lot of the tools to allow that or to facilitate it, we are not the ones who are going to be the primary drivers of it. It’s going to end up being the regulators, it’s going to end up being the primary payers which, that’s Medicare and Medicaid and CMS in the United States, I’d say many of the other groups are just derivative of that. I’m sure there’s people who know way better than I do, but those are conversations that we’re heavily excited about, want to be heavily involved with. I think will change the world and I think if you can seed them with AI that can identify people that are at risk, biomarkers that confirm their stage and would respond to therapeutic intervention, and drugs that actually move the needle, I see a pretty bright future that is attainable.
Ross Katz: That makes a lot of sense and if I may, if I’m intuiting correctly, one of the things that Etiome brings to bear is the opportunity to intervene at diseases much earlier and thereby prevent the progression of further disease and so much of our healthcare payment system, the business models around it, have been built around this idea that you pay to cure a disease that’s already there, you don’t pay to short-circuit a disease that might evolve in the future and so there’s this need to rethink the way that this system works for a world in which it’s possible to do more of this preventative and preemptive medicine. Am I thinking about that right or understanding you correctly?
Scott Lipnick: A million percent. If we can treat people earlier, we can prevent a huge amount of disability from unnecessary hospital visits, unnecessary institutionalization, unnecessarily unnecessary dialysis. End-stage disease costs society huge amounts of money and I think it’s great that we’ve developed the advances to save people’s lives. I just think now we can help people prolong their health. We just have to be really focused on those areas and I think the needs are obviously what I just described, being able to figure out who’s at that stage to be treated, confirm what needs to be treated with and treat them as opposed to saying let’s wait and don’t worry, we can keep you alive. Don’t worry, we can take care of your loved one and make sure that they are not going to suffer anymore. I hope everyone gets to age gracefully. I think part of that is being able to not suffer from chronic conditions that are going to take you from your 60s through your 80s or 90s where you are on extremely expensive, extremely uncomfortable procedures or in situations where you’ve lost a lot of autonomy, whether that’s from cognitive decline or kidney failure or liver failure or what have you. So yeah, I see a future that’s there. I think that if we work together and we have the policy makers and the scientists all aligning and the funders, we can do it. One thing I can tell you is the only group that I interact most with is the Flagship investors and they’re game. They want to make that future possible, they’re investing in it and I think that they’re much more influential than I am and if you look at our leadership they’re out there proselytizing for a better, healthier future. So I think other people will join soon, I joined their mission.
Ross Katz: Yeah, well I can speak for myself and say I certainly share that vision. So as we head toward the end just for people who are intrigued by Etiome and interested in learning more, where should they go?
Scott Lipnick: They should go to etiome.bio. That’s our website. I think it’s really nice. I helped build it. I think the designers who helped build it were awesome so any feedback is welcome. Also the content has a lot of information about what we’re trying to do. Also feel free to reach out to Etiome on LinkedIn or Flagship Pioneering on LinkedIn or you can reach out to me, Scott Lipnick. I’m happy to engage. Would love to meet people who are interested in helping push the mission and vision forward as well as people who just want to give feedback or talk science. We’re also hiring, Flagship both Etiome as well as some other baby companies or protocos that we have in the mix. Would love to hear from anyone. And thanks, Ross.
Ross Katz: Well Scott, it’s been a pleasure having you on today, really appreciate you taking the time and look forward to connecting down the line.
Scott Lipnick: Hey Ross, thank you very much. I really enjoyed this.
Jason: And that’s it for this episode of Data in Biotech. If you enjoyed the episode, please subscribe, rate, or leave a review in your podcast platform of choice. See you next time.
