Panel 4 — COVID-19 Therapeutics | Precision Medicine 2020

[MUSIC PLAYING] [PANEL MEMBERS AND COVID-19 DATA VISUALIZATIONS] ISAAC KOHANE: Good afternoon Welcome back here in Boston, and in the afternoon, to our last session of our sixth annual symposium on patient-driven precision medicine We have a real treat because, as was foreshadowed in several ways throughout the day, there is a lot that we can learn from what’s been happening in both our assessment of and in the treatment of COVID to hyperindividualized therapy And the Principal Deputy Commissioner of the FDA made that very, very clear that she really sees a lot of lessons there and the significant overlap So we’re fortunate in having a star team giving us their view of this And they’ll be introduced shortly I just want to say, in terms of a plea, if you could spare two minutes at the very end when we’re done, I’m going to give just a two-minute synthesis of what I thought were the important messages of the entire day But let’s now proceed with show This panel is being moderated, led, by Professor Galit Alter, who’s a professor of medicine at Harvard Medical School and a group leader at the Ragon Institute of MIT, MGH and Harvard She works on system biology tools to advance therapeutics And she received a bachelor’s and a PhD at McGill and completed her post-doctoral training at the Partners AIDS Center at MGH And clearly has been a leader now, in our Boston-based and international efforts in developing therapeutics So Galit, over to you GALIT ALTER: Thanks so much This is going to be an exciting session I think we’re going to have a fun, interactive, moderated session after the talks But just to begin, I’m going to start off with introducing our first speaker So Dr. Paul Farmer holds an MD and PhD from Harvard University, where he is the chair of the Department of Global Health and Social Medicine He is also the chief of the Division of Global Health Equity at Brigham and Women’s Hospital in Boston Additionally, Dr. Farmer serves as the United Nations special advisor to the Secretary General on community based medicine and lessons from Haiti We are very fortunate to have Paul as our first speaker to kickstart this controversial session Welcome, Paul PAUL FARMER: Thank you, Galit, and Zak, and all of you who organized this for including me I’m going to add by way of explanation to those participating, that I’m an infectious disease doctor, but my PhD is in anthropology And one of the things that I learned as a medical student when I was puzzled about the recommendations

made by specialists which came to resemble very highly patterned, kind of, for example, you consult a neurologist about a patient She might say do an LP Or you consult a gastroenterologist about a patient and she might recommend endoscopy And with a patient that had asked not to have an endoscopy, a patient who was moribund– I was still a med student– I asked the resident, well, why did this patient have a endoscopy? The patient died a few hours later And the resident said to me, well, you called a GI consult, a gastroenterology consult And I said, yeah He said, ask a pizza man what’s for dinner, he’ll tell you pizza So anthropologists, medical anthropologists, are going to talk about social, the social aspects of precision medicine And going back to the definition, one of the definitions that appears in the materials for today, which I’m sure was already discussed earlier this morning, from the National Academy of Sciences committee, meant taking an explicit multi-dimensional view of patients, not just one data modality, such as genomics or environmental exposure So subjecting the current crisis, the COVID crisis, because there are several going on at once, what are the multi-dimensional views that we’re often missing? And I’ll say this as someone who’s been involved, clinically, and also in evaluating the contacts of people in Massachusetts who have been diagnosed with COVID And what we see is that the multi-dimensional view that many adopt in medicine and public health erases or elides the social conditions that determine, surely, risk of infection, or determine much of risk of infection, and also determine the course to a lesser, sometimes, but invariably, to the course of the illness And let me give some examples And I would argue that as much as I’m down with the term precision medicine, social medicine, which is a much older construct, a couple hundred years older, probably, also calls for multi-dimensional views of patients And many in social medicine are very interested in epigenetics and are obliged to be interested in environmental risks So my argument here, just to launch the controversy, if that’s what Galit wants, is to argue that without a broader view, which precision medicine can offer, and social medicine reliably does, without a broader view, we run the risk of impoverishing our analysis of what’s happening with COVID in individual patients, in communities, and in countries with histories as varied not only of Taiwan and the United States, but Italy and Germany So how do we explain such massive variation in spread and in case fatality? And a truly comprehensive view of precision medicine would answer that challenge You will remember, probably in January, when we first started reading– some people maybe earlier– first started reading about a new pathogen that caused a viral pneumonia, there were comments about, well, this could be the big one, which it seems to be becoming, but also erroneously claims that this was respiratory pathogen, it was going to spread everywhere, no one had immunity to this pathogen, it was alleged, therefore, would become a great leveler That is that we shared risk as a species because we did not have antibodies to the novel coronavirus SAMANTHA: One minute, Paul PAUL FARMER: Oops One minute? Well, all this to say that we now know that that was completely false That the color of COVID is brown and black, that the course of illness and the risk for infection both have been greater in communities of color In some cohorts, we saw the majority of patients presenting with– in some cohorts, urban cohorts, majority of patients presenting were people of color, when they were not the majority of the city or polity in which they were found Instead, we found prisons, packing plants, and of course, nursing homes as the built environment, in other words, as amplifiers of the epidemic Now, in the 19th century– and I promise I’ll close–

many of these variations in risk and outcome would have been attributed to genetic or hereditary, what were called hereditary factors Those usually have not panned out, and certainly in the major discussions of, for example, why do so many people of color– that was not the term at the time– why did they die of tuberculosis at this disproportionate rates? Those did not pan out much And epigenetics are being studied now But in closing, what I would offer as my challenge– and I’m going to learn a lot here– is that to have truly precision medicine, we are not going to be able to look at the pathogen to describe its radically varied course in different populations As this conference calls for, these are patient-driven or host factors, very reliably, and they’re socially constructed And I will ask, leave it at that And I will close in saying that we really do need to look at conditions that are not readily, they’re not often even studied in medicine How do social inequalities arise? How do they become sustained? And how do they alter risk of infection and of the course of illness? So thank you for allowing me to, you know, to insert that view in this conference GALIT ALTER: Thanks, Paul That was wonderful So I think we’re going from population to social structure as a way to think about how we’re going to deploy therapeutics and vaccines So our next speaker is Albert Barabási, who is the Robert Gray Dodge Professor of Network Science and a Distinguished University Professor at Northeastern University, where he directs the Center for Complex Network Research and holds appointments in the departments of Physics and Computer Science, as well as in the Department of Medicine at Harvard Medical School and Brigham and Women’s Hospital He is a member of the Center for Cancer Systems Biology at the Dana Farber Cancer Institute He’s Hungarian-born, a native of Transylvania, Romania, and he received his degrees from Budapest Dr. Barabási, can you take control of the screen? Welcome, and we look forward to this very different perspective on precision medicine in the context of this discussion ALBERT-LASZLO BARABASI: Indeed, Galit And thanks, Galit and Isaac for having me And I define myself as a network scientist And I spent the last 10 years really thinking about how we can actually use network-based tools to really predict the drugs, and predict the impact of the drug, and eventually to design new drugs And in that process, we ended up kind of developing a series of tools that I will share very briefly with you, effectively trying to kind of detect potential drug targets in different complex systems, different complex cellular systems And so the idea behind network medicine that we pursue in my lab– and I’m assuming that you see my slides now– is that nothing in the cell that happens, happens in isolation It happens to interconnectedness The molecules interact with each other They bind to each other They engage in reactions with each other and so on And you must account for that interconnectivity in order to really think about how a disease emerges, and eventually, how do we cure it, and how do we intervene to cure it? And the tools of network medicine allow us actually to define the disease module, which the network neighborhood or the disease incites Think of the cell as being the map of Boston, and you would say a certain disease only resides in a certain network neighborhood So the disease module is that neighborhood But once we have that disease module, we can actually identify drugs that hit in that particular neighborhood and have the potential to perturb the disease as well as drug combinations And eventually, this will lead us, does lead us to what we call personalized network medicine, or what you would call precision medicine, where we use combinations of drugs to be really [INAUDIBLE] individual Now, when the COVID hit, we, as many others, were really motivated to do something about it And we realized that we also had the moral responsibility, because these tools were really designed to rapidly identify drug repurposing candidates, and we needed drugs, and we need drugs So we deployed these tools to really understand what drugs could be effective against COVID-19 And to understand how we approached that, first, we are starting from the picture we have We are curating all the physical interactions

that are existing in the cell And that’s what I call the human interactome here on my slide These are typically physical interactions between human proteins But then, of course, this human interactome is being, now, perturbed by the virus, because the viral proteins enter and then bind to some certain human proteins And as soon as the data became available, which human proteins this binds to, the viral proteins bind to, then, we were able to define what we called the COVID-19 disease module, which is the network neighborhood within the human cell where that is being perturbed, or hijacked by the virus And the question we need to ask in the context of drug design or drug repurposing is that, are there already drugs on the market that hit in the right neighborhood, so that they could actually perturb the ability of the virus to bind? And so in order to execute this program, we start, of course, from the human interactome map, which about 8,000 proteins and between them, about 300,000 interactions We start from the targets of the SARS-CoV virus, which about 320 human proteins And of course, we start with a drug target list for about 7,500 approved drugs And what we did is that we deployed three different methodologies that were independently developed by other groups and us to identify drug repurposing candidates Some of them are based on what we call network proximity, simply who hits close at our COVID targets? Others, however, are based on AI methodologies that look at the world picture and predict who could be actually the proper drug repurposing candidates SAMANTHA: One minute, Laszlo ALBERT-LASZLO BARABASI: Sure And the result of that was a very long list of drugs, a prioritized list of drugs from which our colleague, Joe Loscalzo has selected 86 that are clinically potentially relevant So what we did was if we took all the 7,000 drugs that are out there and we rank them for their potential ability to perturb the COVID disease Of which 86 now are sent to actually needle, and they were tested experimentally both in monkey cells and the human cells And just to give you a reference, when you actually test, typically, a recent test about 12,000 drugs had a 0.25% hit list, hit rate However, the list that you see in front of us, 47%, according to the experimental data, had an effect on the COVID So now, these are being transferred to human cells, being tested once again there, and hopefully in a week or so, we’ll be able to do clinical trials on them Why am I showing you all of this? Not because to look at the particular drugs, but to say, we have a very rapid methodology now to test any drug for its potential relevance for COVID And once we have tested high, we can also tell you the disease mechanism How is actually that particular drug is perturbing the COVID module? And I think this will be a model, not only for COVID, but also future diseases for which we can very quickly screen the existing drugs for potential impact Thank you GALIT ALTER: Thank you very much That was wonderful Right down to the atomic level of precision OK, so our next speaker is Dr. Russ Altman, who is the Kenneth Fong Professor of Bioengineering, Genetics, Medicine, Biomedical Data Science, and Computer Science, and past chairman of the Bioengineering Department at Stanford University His primary research interests are on the application of computing or artificial intelligence to problems relevant to medicine He’s particularly interested in methods of understanding drug action at the molecular, cellular, organismal, and population level So I think we’re going to get the full picture here So if you can grab control of the screen, Russ, welcome, and we look forward to your seven minutes of controversy RUSS ALTMAN: Thank you And it’s great, and thanks the organizers for inviting me, and best wishes to everybody So I want to do a couple of disclosures I am a general internist, MD But I am not currently treating COVID patients So my instincts about hyperindividualized therapy will not be coming from being in the trenches It will be coming from thinking about the trenches from, basically this room, for the last three months Secondly, I am a biomedical informatician, data scientist, and AI person And that informs a lot of what I’m about to say And I’d like to say that, initially, when I was invited to a session on hyperindividualized treatments for COVID, I laughed because I would take a totally generic treatment for COVID-19

right now And so the idea of hyperindividualizing– but however, in the context, especially of Paul Farmer’s comments, when you think about the broader picture, there are actually many things we can do in a hyperindividualized way at the social and cultural levels that I do recognize The final disclosure is, as a data person, I feel very lucky because we can hop around to different scales And in fact, Galit anticipated that I would be a little bit like this When you are a data person, you are not wed to any particular experimental scale So whereas there is a real fundamental question about whether we are real scientists or not because we don’t go and do experiments in general That aside, we have students, and postdocs, and staff, shoulder-to-shoulder, some of whom are looking at the molecular level, some of whom are looking at claims data, some of whom are looking at global data And you can do that when it’s all just data And I think that’s the power of the field OK So now, I’m going to share slides We, I think it’s clear to say, the great variation in COVID-19 disease is not fully understood, but we have amazing leads But before we even look at data, we know some categories We know that there may be variations in the virus that we need to pay attention to that would lead to different outcomes And again, building on what Paul said, we know that there are environmental– and this is environmental writ large: cultural, social, behavioral, and exposures– that could be altering the response to the viral exposure There are many of our colleagues looking at the genetic efforts The UK Biobank released some of their genetic data for the patients who had been infected by COVID and I think some controls And there are others as well So before you even look at data, you know that there are these bins And they all require confirmation, ideally, a mechanism, talk about that in a second, and intervention options In general, data will give us correlations Sometimes data gives us causation, but it is a much more complicated activity But those correlations still set us on a path And in general, I call it, when we’re trying to understand mechanisms, I call that science And so data is a hypothesis generator, at least And then we use data and other mechanisms to understand the mechanisms So the task is to look at correlations and mechanisms to try to understand what we might be able to focus on in an individual in order to give them the best chance at success in going through with this disease So this is my final slide And I want to tell you about three activities that we’re doing that really do go across a range of scales We’re doing collaboratively with UnitedHealth Group, they cover something like 90 million lives, maybe more than that, in the United States And therefore, a large fraction of patients who have had COVID have their records, both claims and lab records, available to United And to United’s credit, they offered to work with us to look at the drugs that patients were on, and try to understand if there was any evidence in their data about drugs that might be risky or protective Now, what do I mean by that? If we saw that patients with COVID in the hospital, compared to patients with COVID who are not in the hospital, were taking a drug more frequently as an association or a correlation, that would be interesting Because that would tell us that either that drug is causing problems that leads to hospitalization, or that that drug is correlated with something that might lead to the hospitalization– SAMANTHA: One minute RUSS ALTMAN: Such as the disease that it treats Did I hear one minute? SAMANTHA: One minute RUSS ALTMAN: Yes? And then, also, even more interesting, what happens if the people in the hospital are, on average, less likely to be taking a drug that the people who are out in the community with COVID, but not hospitalized, are taking? That might be a protective correlation or hypothesis So I can tell you that we have found both risky and protective drugs We are not convinced yet that our controls and our statistics are right, but I hope to be helping UnitedHealth Group report that in the future And this would be a source, potentially, of a hyperindividualized therapy based on what the diseases are and what’s being treated I will skip the second one because it’s very similar to what Dr. Barabási just said And I’ll go to the third one, which is that in collaboration, in an international collaboration with the University of Western Australia, and folks there who have long relationships with Jakarta and Indonesia, we are looking at trials for early intervention

So as soon as you get tested positive, can we put you on a cocktail of drugs to alter the course of the disease in a meaningful way? Can we monitor you at home, and can we generate using drugs such as Dr. Barabási just disclosed, and also some ones that we have some ideas about, can we create drug cocktails that bend the curve, not for the population, but for the individual patient? If we can do that, that would be a big deal And there is a very good clinical trials infrastructure there, there are good relationships And it’s still in the planning phases, but it’s an exciting thing to be part of because it might lead to some useful answers All of this, I must say, while we await new treatments and new vaccines, which of course, must be the name of the game But until then, I think that these kinds of efforts for repurposing and understanding what the data is telling us about drugs to which people have already been exposed is an important opportunity So thanks very much I look forward to the discussion GALIT ALTER: Thanks very much So our next speaker is going to be Dr. Mark Namchuk Dr. Mark Namchuk held a number of senior research positions over a 17-year career at Vertex Pharmaceuticals, including SVP of North American Research and interim Global Head of Research Mark started his drug discovery career at Cubist as the head of the enzymology group Over the last 24 years, Mark has directed drug discovery efforts in numerous therapeutic areas, including infectious disease, oncology, neurodegenerative disease, and psychiatric disorders Mark recently joined Harvard Medical School in 2020 as the Executive Director of Therapeutics and Translation, just in time for the COVID epidemic We are now fortunate to have Mark close off this particular session, or this part of the session, so we can get into the fun discussion component Mark, can you grab the screen and take into the debate? MARK NAMCHUK: Great First of all, thanks to the organizers, Zak, in particular, for the invitation to participate I’m going to give you a slightly different view, that I think, in infectious disease, one of the unique challenges, particularly if it’s an infectious disease that you want to be able to treat on a worldwide scale, the precision can’t be in understanding the patient response, the precision has to be in the molecule And in fact, if one takes a step back and asks, what’s historically been the most successful way to treat a virus, it has been to design molecules specifically to go after viral targets that are expressed by that virus And then, to be able to remove anything else that it would be hitting from a host or otherwise perspective So I think that one of the challenges that has come across in the recent crisis has been, the entire world stepped away from infectious disease research for a period of time And in particular, away from the research that would be aimed at developing medicines specifically to treat a virus So I’m going to share my screen here And this is a slide that we’ve been using for the therapeutics working group within the Massachusetts Consortium for Pathogen Readiness It’s a group that both myself and Galit belong to And I think it’s emblematic of the challenge that’s in front of us And I think I’ll seize on something that Dr. Altman said I think you want to do something that you can do today, but I think we have to acknowledge our probability of success goes up as these efforts become more constrained and focused down on the actual virus that we want to treat So without doubt, and a big part of what we’ve been doing in the Mass CPR Consortium is to look for the potential to repurpose drugs I think if we look at our success to date in the COVID outbreak, it’s worth noting that the only molecule that’s shown credible clinical activity was actually technically not repurposed It was an antiviral designed to be broadly acting on the RNA polymerase of an RNA drug And that, to date, other things that have been pulled in from other indications have shown modest success, or in some cases no data to support efficacy to date Now, the methods are getting better And I think the things that are going to be coming in the future are much more likely to work But I think we also have to understand, as a lifelong drug design guy who used to work on structure-based design and chemistry, the way I think about building a drug, is it’s a little like tailoring a suit And let’s imagine that the person we’re tailoring for is six-foot-four, broad shoulders, no waist, big guy And you’ve worked on this jacket for five years And then you’re going to go ask whether that jacket fits someone else It might But the likelihood that that jacket’s going to fit again is not nearly as good is going to tailor another jacket

So two of the other activities that we’ve invested pretty heavily within the consortium, and which I think have to be pursued in parallel with drug repurposing efforts, are things like therapeutic antibodies where, again, you’re engineering something to have an exquisite amount of selectivity to just treat the virus And then, also, to broaden out what we can think about from a repurposing perspective into molecules that don’t have as much clinical experience yet, but as we gain better understanding of the replicative cycle for the virus, have a more likely mechanism to bump into what would actually drive replication Maybe my closing remark would be to say that, as we think about not ending up in the situation we’re in right now with a corona outbreak in the future, one of the things that we can borrow from flu preparedness is to begin to design drugs that are not only going to be useful for the current outbreak, but steel our understanding of zoonotic transmission, genetic drift, to try and design molecules that actually treat corona, but with a broader spectrum of action So a beautiful amount of imprecision, if I will, to be the skunk at the picnic on precise medicine, and that in fact, one of the issues that has really befallen us in the coronavirus outbreak, there is nothing from a therapeutic perspective that was ready and off the shelf that could be the first line of defense as we’ve begun to rally an extraordinary effort around either a new therapy from repurposing or a vaccine And I think we’re now living through the consequences, both social, economic, and medical, of not having made that sort of an investment When I was still at Vertex, one of the programs that I ran was in pandemic flu And in fact, a molecule that we developed is now in phase three testing for influenza And I think at that time, most of the preparedness agencies saw therapy, not just through the window of being able to interdict when someone was already ill, but really through that window of the first line of defense So my advocacy as we think about precision would be to remember that we have to do our chemistry very precisely to prepare for the next wave of corona outbreaks We are now, unfortunately, potentially in an era where we may see a zoonotic transmission every six years with varying amounts of lethality and varying amounts of contagiousness And that while we should rally to what we can do immediately, that we should place efforts into using what we’ve learned over 30 years of doing design of antivirals to be more prepared for the next time Thank you GALIT ALTER: That was great That was great Thank you, Mark So now, we’re going to go into a bit of a discussion for a few minutes And so I’m glad everyone’s up So maybe I’ll start off by asking the first question So just as a counter argument to Mark’s last point, the idea of going after the molecule, precision molecular design as opposed to precision medical design, is there not a role for the way that Dr Barabási’s thinking about the situation, as opposed to thinking only about the viral target, but thinking about all the other molecules in a given individual that can shape how that one target molecule functions within a given individual, is that not pertinent? And thinking about all those genetic variants that could influence how that molecule is targeted, is that not where precision medicine at an individual level does play into efficacy of molecular design of drugs? [INTERPOSING VOICES] GALIT ALTER: Either one, go for it ALBERT-LASZLO BARABASI: Then I’m happy MARK NAMCHUK: I’ll go second [LAUGHS] ALBERT-LASZLO BARABASI: Oh, I see, OK Well, I mean, it is And I think, but I think we’re also have to look at, when we think about precision medicine as a time scale issue, right? I think if we have all the time available and we would have a fast way of approving drugs, then Mark’s way would be the perfect way to go, and that’s what we should be doing The reason we actually kind of embarked on the repurposing is because it was pretty clear it was impossible to get at the timescale we needed the new drug on the market It’s just not going to hit the market at that timescale And so therefore, I totally agree with the analogy that I would love to have a suit that is perfectly tailored to me But if I were to be the hospital, I’d be happy with anything that covers me and gets me out of the hospital alive Right? So I think of drug repurposing, indeed, as not getting the right suit, but getting any clothes that would cover me and protect me for the time being And that’s what we can actually achieve to the drug repurposing process And this can actually go further So one of the things we’re now starting to do in the lab is, as Russ has mentioned, actually, is effectively to do, now, drug combinations The same pipeline that can predict individual drugs

can predict drug combinations as well And also, as micro data will start emerging above the individual patients, we may have the opportunity to tailor to individual patients to drugs Well, for that, we need an arsenal We need something to throw at the patients And I think we’re still in the first phase of figuring out, what are the keyboards that we can touch? And once we have the keyboards, then we can push them together, and that’s the next phase GALIT ALTER: Great, I like that Tailor-made drugs later MARK NAMCHUK: [LAUGHS] So I’ll just, I’ll say that I actually completely agree with that point of view I’ll only add two points, Galit, maybe to the way you had phrased the question, I think that one of the challenges with an antiviral or an antibiotic that you want to deploy worldwide is you actually hope to design something that does not depend on the genetic predeterminants And in particular, if it’s going to be deployed in environments where you don’t have access to that sort of molecular precision, that’s going to be a detriment to its broad utility So I think that when you can do it– and I think particularly with SARS-CoV-2 in terms of the severe inflammatory syndrome, I think that is what we will have to do We’ll have to chop the disease up into pieces and understand why individual patients respond differently GALIT ALTER: [INAUDIBLE] MARK NAMCHUK: But as a goal– GALIT ALTER: Yeah MARK NAMCHUK: But as a goal, I think to make something broadly deployable, you need to go to the highest common denominator of the biology, in which my argument would be, is the virus itself, and what makes virology and bacteriology different from other areas of medicine GALIT ALTER: For a second, I thought you sounded like an academic MARK NAMCHUK: Just a second GALIT ALTER: Just a second OK Let me go to Paul, and I want to come back to Russ So Paul, I’ve got a question for you I know you started talking about, thinking about different populations, and you really, like, kind of struck a chord there So have we not been doing precision medicine already, tailored to different populations, as we go into different socioeconomic categories and different ethnic groups? When we think about what is acceptable therapeutically in one group versus another, or what might work from a perception perspective or from a genetic perspective, have we not been doing this already in some shape or form? And how can we start thinking about precision population level design, because that’s already going on? You’re on mute OK, there, you’re not anymore PAUL FARMER: All right First of all, I think it’s clear, I hope it’s clear, rather, that as a clinician, I’m banking on my colleagues here who are looking for precision medicine at the molecular level And looking back at some of the– if you take viral pathogens that cause enormous mortality– HIV, HCV– that has clearly been the recipe for a great deal of success, right? The strategies that, as Mark said, have been used over the last 30 years And then, once we have an effective deliverable, and by that, I mean a vaccine for a preventive, vaccine or other preventive, a diagnostic, and these are also going to be molecular And as we know from our colleagues, there’s very powerful ways to understand the nature of community transmission through genetic sequences So again, sorry, preventive, diagnostic, and of course, finally therapeutic So I’m counting on you all to find the deliverables, which are going to come out of labs and out of the efforts of Dr. Barabási, who has described looking at already available somewhere in the market So the population-based, just to twist your question, Galit, yeah, we are delivering precision-based medicine, but it’s not precision-based medicine that is lessening the cull of COVID-19 among communities marginalized by some of the forces that we have So as a team player who is counting on the trialists and the basic scientists to develop that molecular-level precision and that systems-level precision in an individual patient, there are clearly things we could do to render our social diagnosis and social support much more precise, if you will And if you’ll just give me one example So if we’re doing a Massachusetts-wide contact tracing initiative– and just to give you some of the numbers as I understand them, of the 100,000 or so cases in Massachusetts, the team that we’re working with has already been in contact with 26,000 of them, and made 300,000 calls

to their contacts And that required hiring 1,900 people When the contact tracers talk on the phone to either those already diagnosed or those who are contacts of the diagnosed, what we hear is, often, is a lot of talk about their social conditions Whether or not they’re essential workers who are obliged to take risks that people who have other, more resources, don’t take Whether they have comorbid disease like diabetes, or COPD, whatever it may be, whether they have alterations in renal or hepatic function So I do think that we’re doing it, Galit, but I think we’re doing it the wrong way We’re not tailoring our multi-dimensional view of precision medicine to the whole individual, and certainly not to entire communities yet The good thing is that can be done GALIT ALTER: Well, hopefully we’ll learn Yeah PAUL FARMER: That can be done GALIT ALTER: Thank you That’s great, Paul Well hopefully we’ll learn OK, so one last question I have for Russ So one question is about testing drugs And so right now, we’ve got all these drugs that are flooding the market going into very rapid testing within hospitals, and all the patients trying to get something out there as fast as possible Shouldn’t we have been binning already based on all of your categories that clearly are emerging? Are we doing this initial discovery piece properly as opposed to only thinking about what’s happening between the mild and the severe disease individuals? And can we be using this type of AI information to do precision-level clinical trial testing to get us that end goal more effectively? You’re on mute RUSS ALTMAN: Great So thank you very much for that question And yes, so there’s so many factors in your question, so let me just try to pick out a couple So first of all, I think there are opportunities for stratifying patients in rational ways that will make the drug trials more effective and convincing, because just giving everybody in the population a drug is going to be risky, and risks no result So that was a great point The second very important point is, we are looking at novel trial designs You can almost imagine giving, like, almost like a binary search Like, we give two medications to a bunch of arms, and then depending on how those arms do, we either merge the arms, or separate out the arms And so there are very talented clinical trial design people who can apply that So you can use the AI on the front end to divide up the patients You can also use the AI in the trial design You also talked about the monitoring So just deciding what to follow in these patients, and what a measure of success is One big argument we had is we had some people saying we have to follow the viral RNA And there were other people saying, no, we have to follow clinical outcomes And of course we care about clinical outcomes, but the viral RNA designs led to a much more efficient trial, but then you’re asking community health leaders to trust that as a marker for likely health outcomes, but not proof of a health outcome So I think the high level is, you’re right We can step back, now, and try to be very rational The population of patients for these trials is, at least in the Bay Area, around San Francisco and Stanford area, we have been pretty successful at lowering the curve That’s great for our patients and for our community, but it means that our trials have all come to a standstill If you say, Russ, why are you talking about Jakarta and Indonesia? They’re expecting a big pulse Ramadan is over There’s a bunch of reasons why, unfortunately, we might see a spike And so we’re looking around for the right places to conduct these trials Of course, with appropriate ethics and appropriate on-the-ground clinical care so that we can get these answers And in the Bay Area, there are literally tens and tens of trials that have just stalled because of recruitment So it’s a huge issue GALIT ALTER: Absolutely Yeah So hopefully we can add precision medicine to the front end of the drug development, not just to the back end RUSS ALTMAN: That’s right GALIT ALTER: OK, thank you very much This has been a wonderful panel Thank you all for your willingness to go rapid fire And with that, I’m going to turn it over to Zak ISAAC KOHANE: Me Yes GALIT ALTER: OK, thank you ISAAC KOHANE: All right Galit, that was wonderful You should have taken over the whole conference It would have been somewhat easier for me, and more entertaining So this is the portion of the panel where we invite audience questions that will be read out loud to us by Dr. Ken Mandl But as before, we have people learn how to use their Slido interface by entering, by seeing a poll, and then the quiz As foreshadowed before, there will

be three very nice branded prizes made to the three top winners of our quiz And those will be posted later on our website So congratulations in advance to whomever you are And I actually said the answer to this question in a previous panel, so you should know the answer to that Man Apparently communication is not my forte OK RACHEL: And maybe remind folks that the Slido is right there on the same web page, if you just scroll down below the live stream ISAAC KOHANE: That’s right That’s right All right RACHEL: So we’ll just give this another 10 seconds or so before we move on to the quiz ISAAC KOHANE: Yep Let’s go They listened Well, all righty, then So Dr. Mandl, I presume you’re seeing a list of questions from the audience Can you please proceed? KEN MANDL: Great So let me start, actually, by asking a question to Paul And Paul, your pizza man anecdote at the beginning reminds me of the exact analogy I had on one of my first days on the wards as an intern, where the same thing, a GI consult, a obligatory endoscopy, but the comment back was, you go to Midas, you get a muffler And Zak, that was, as you will recall, from the great Esau Simmons, once upon a time So– ISAAC KOHANE: You could conclude that GI is a fount of great anecdotes KEN MANDL: It seems to be a theme So Paul, if we were to roll back the clock to, let’s say, March, and you are now in charge, prospectively, of precision social medicine, what’s your prescription for preventing Chelsea, Massachusetts from happening? PAUL FARMER: I wish I had prepared for that question It’s such a perfect one, and one that everybody who is a critic should, and involved in the response, should be obliged to answer But I can say some of the things that I did say Some of us argued, especially when we heard this specious talk about the great leveler, that we should really concentrate on getting ready to do contact tracing and to provide, because testing, tracing, and then, supportive isolation are the next part What does supportive isolation mean? It means having a place to go where you’re being fed, and cared for, and sheltered, and not in contact with anyone Which I’m able to do here, but most people who I see as patients or are not able to do That sounds very vague, but it’s not Because we’ve been doing it later We’ve been renting hotels, we’ve been, Harvard University it’s certainly going to have The Inn at Harvard at its disposal, and many other buildings no doubt So that’s something that I wish we had done early We also recommended large scale amnesty for prisoners And had we thought about meat packing plants, we would have said, well, there’s no possibility of having social distancing here And again, all this while waiting for the specific therapies that we’re discussing today, whether they’re, or preventive So I think there were a lot of things that could be done I, for the life of me, all of us in social medicine, and infectious disease, and epidemiology, et cetera, and precision medicine, are struggling with the radically varied case fatality rates I mean, these are not going to be described you know through genetic differences between Italians and Germans or between people in Taiwan or Wuhan

So I do think, I mean, that’s kind of a lame list, but they’re all actionable policy items And in fact, they’re being embraced now tardily after you know a widespread failure to prevent the massive amount of deaths we’ve seen in American cities Is that at least a decent answer? KEN MANDL: That’s great Thank you, Paul PAUL FARMER: You’ll inspire me to write something that answers that very lucidly and concisely, I mean it KEN MANDL: I would love to hear your further thoughts as well So let me ask a question to Mark and Russ to tag team this one from the audience How would you advise researchers and drug makers to identify causality using ML and AI, rather than mere correlation? So this may begin with a explanation of what deep learning is, but not too long, please MARK NAMCHUK: I think Russ should kick that one off, for sure [LAUGHS] RUSS ALTMAN: OK, so ML is machine learning Deep learning is a type of machine learning It has two main features It is capable of amazing performance and it is virtually inscrutable in terms of understanding why the good performance happens Those are all generalizations that I will press submit on as I’ve been taught So it is possible there is a whole theory of AI and causal reasoning different from correlation Everybody knows that correlation may or may not be because of causation There is a higher bar for causation when analyzing data, and there are theories about how to do this I would say that, at a time of emergency where those theories are still emerging, that it would be fine to try them, but you need to use the old fashioned methods for mechanism today And that’s experimental science with appropriate controls In the clinical sphere, it’s randomized controlled trials, and in the molecular sphere, it’s the hard work of viral assays, isolating your systems, and testing the impact of the molecules on these systems So I am not bullish in June of 2020 about a big causal reasoning chain for AI I don’t rule it out as a possibility, but I don’t think we should bank on it because these are emerging ideas that are very new MARK NAMCHUK: Yeah, I think the only thing I would add is– and this is maybe stepping away from the computational pieces of it, which the other people on the panel are way more qualified to comment on than I am But I think one of the things, and when you think about application of genetics in medicine, where we’re still challenged is we find an association, genetically, we don’t know if one intervenes at the time of disease diagnosis with that gene, whether it’s interventionally useful So expanding that back to COVID, I think what a lot of the broader things, like Dr. Barabási’s work, et cetera so, is they create this big pot of ideas that we can use And that’s great I think what you then need to filter down is, which of these are the ones that make sense to our growing understanding of the biology? Maybe those go to the top Maybe you have to be a little biased based on that information coming from the biology, and then test empirically, in that sort of an order But also, understand that sometimes what we get are associations that are not necessarily telling us what would happen and be useful interventionally And I think the biggest challenge in SARS-CoV-2 has been, it is probable, but we don’t know that if you intervene too late with something that’s only antiviral, it may not do anything to alter the disease course once the immune components have taken over So there, I think understanding that transition, understanding that at a deep level, being able to use techniques like AI and machine learning to build when that crossover point might be, but to build a big pot of ideas is probably the right way to prioritize what you take into empirical testing KEN MANDL: So let me see if Dr. Barbaris, sorry, Barabási would like to add to this conversation which is actually going at a good clip right now ALBERT-LASZLO BARABASI: Sure And this is a very good question Of course, at the end the clinical trials are the ones that tell us whether there is really causation there, right? That’s what they’re for I think a bigger challenge for us is really, first, is to emphasize that these methodologies are about identifying drugs that can perturb the right neighborhood of the network And it could actually make it worse, rather than improving the disease state So that’s where the experimental testing and the clinical trial is necessary For us, actually, in the machine learning and the AI space, one of the very interesting questions was, why does the algorithm make that prediction? And we spent quite a bit of time, actually, extracting what is the information these algorithms are using to make these specific predictions

and how they differ, and how they not And if I want to step back at the big picture, the way I look at it is that, let’s face it, we may not be able to do something really useful in the time space is necessary But COVID, for us, is becoming the new E. coli That is, that becoming the system where there is so much data available that spans everything from the molecular mechanisms all the way to the social effects, and contact tracing, and everything, that it will become a unique sys– GALIT ALTER: Volume ISAAC KOHANE: The Hungarian self-censoring algorithm broke in KEN MANDL: So– ALBERT-LASZLO BARABASI: Evolving right now for our scientific community And I think that’s the value, and that’s why we have to continue working for it, even if we miss the opportunity to have patients to treat And I’m hoping that by the time these drugs are ready, there will be no one to treat with them KEN MANDL: Great I have a good question to, for either an ultimate or a penultimate question Let me actually have Zak take the first crack at this one, but then open it up to this whole group, who should have– SAMANTHA: Ken, we’re actually just out of time So if we could just do final comments, close it up KEN MANDL: No last question? SAMANTHA: We could do this one last question, but I don’t think we’ll be able to open up to the whole group for comment KEN MANDL: OK [INTERPOSING VOICES] KEN MANDL: This is from a student At present, there is no direct program for studying precision medicine For a medical student outside of the US, what would be the best path to research, develop, and practice in the field of hyperindividualized medicine? ISAAC KOHANE: Wow, that is such a great question And I don’t have a good answer to it Because it’s really a multidisciplinary approach And very few medical schools, very few universities, have courses which bring it all together As Paul points out, if you don’t involve the social aspect of it, you’re missing out If you don’t include the reimbursement part of it, you miss out As we learned from Julia, when she talked about her kid with– Mila– she said, the problem is not the science They figured that was easy Even getting money for the therapy was not that hard It was actually finding Tim Yu who would actually do it turned out to be hard and important And what we’re hearing from Amy Abernethy at the FDA is that we need these new infrastructures, these new platforms to be able to bring these things together And what I hear from Mark is that maybe some of the academic approaches don’t even work necessarily particularly well And so I think that, collectively, what we see here is a précis of many of the disciplines that need to come together to create a hyperindividualized medicine And I hope that this conference, in its multiple instantiations, is a goad towards getting there So let me reassure my colleague from a not US university that there are essentially very few universities, unless the annoying Stanford representative wants to disagree with me I just don’t think this is taught systematically in this country, either So I think it’s a multidisciplinary opportunity And I want to point out that when William Welch heard about what was going on in Europe about understanding infectious diseases, having spent time in Vienna and seeing what was done with handwashing, which, still, we don’t do that well until very recently, and learning about infectious diseases, what he was able to do was transform American medicine into a much more data-driven, scientifically-driven process in a way that medicine then was not And as a result, in the Flexner Report, that’s something that happened in 1910, half the American medical schools closed down Some unfortunately for some really underserved communities [INTERPOSING VOICES] SAMANTHA: We have one minute left ISAAC KOHANE: Yeah But the point is that these communities, that this transformation is something that does have to happen And I think this conference highlights it So I just want to really end there, and thank you

all for your participation I want to thank our sponsors, Merck and Metadata I want to thank our many speakers, as well as this panel, and moderators, such as Ken Mandl, Raj Manrai and Matt Might And most importantly, the people behind the scenes This is much harder than it should be Harlan Reiniger, Catherine Ward, Rachel Eastwood, and most importantly, Samantha Lemos, who’s really steered the whole thing Thank you very much [APPLAUSE] And with that, I hope the weather is good wherever you are and you can enjoy a bit of the day Thank you all See you next year, hopefully in the flesh [MUSIC PLAYING]