The New Face of Personalized Medicine

Moderator: Kevin Davies, PhD Panelists: Colin Hill, CEO GNS Healthcare Eric Shadt, PhD, CSO Pacific Biosciences, Chairman of genetics at Mount Sinai Medical School Felix Frueh, PhD, President, Medco Research Institute Kevin Davies: It is my great pleasure to be invited to host this one-hour webcast which should be a fascinating discussion. My thanks to you for a second time and tuning in and to GNS Healthcare for their assistance in putting on this interesting panel. I'm going to introduce our three guests in just a minute. They are, however, Eric Schadt from Pacific Biosciences and the Mount Sinai School of Medicine, Felix Frueh from Medco Health Solutions, and Colin Hill from GNS Healthcare. A bit more about them and Ill have them also add to that introduction in just a minute. This is an interactive discussion. So we really look forward to your questions, and we'll try to include as many as we can during the course of the one-hour discussion. You can enter them directly on the dialogue box that you should see on your computer screen. Or if you are interested, you are invited to send them using the #pmchat-- PM, personalized medicine chat-- which my colleagues would be following. And they'll relay any provocative, interesting questions to me which we'll be take notes on the second half of our discussion. So is this just another discussion on personalized medicine? Well, yes and no. Let me just set the stage, and we'll bring in our panelists. We've all seen, and indeed, many of you have contributed to the stunning progress in technology development and biomedical research. And thanks to that, we've got more preventive, diagnostic and treatments for human disease than ever before. Targeted therapies, companion diagnostics, we have the arrival of the last 12 months or so whole genome sequencing into the clinic with very positive impact. And we're learning more each day about the allimportant interpretation of the human genome thanks to a plethora of new software tools and other things. But we still really haven't successfully translated that in any kind of meaningful way into the world of healthcare. We've seen the arrival of big data in terms of omics and systems biology, which give us the tools to make headway in developing personalized effective medicines. But how will they be translated on a huge or massive scale? What does the arrival of the $1,000 genome only a year or two away mean for diagnostics and personalized medicine, drug discovery, or health care? And has this happened before, being hoodwinked by the razzle dazzle of this wonderful world of genomics and next generation sequencing? Or should we be thinking in terms of implementing genuine personalized medicine? Should we be looking through other types of omics, and computer modeling technologies, indeed other forms of big data analytics? My three guests today are all distinguished scientists whose expertise runs the gamut from basic research in academia, and nonprofits, and big pharma to pharmaceutical drug development, pharmacogenomics, and the all important pharma-patient healthcare interface. Together they're going to look at the new face of personalized medicine. And we hope to figure out some ideas and solutions to how the industry can achieve targeted

disease discovery that leads to better treatments that work for all stakeholders. So we'll get all that done in the next 60 minutes. Eric Schadt is the chief scientific officer of Pacific Biosciences, recently appointed to chairman of a brand new and very exciting institute for genomics and multi-scale biology, a term I believe he coined, at the Mount Sinai School of Medicine in New York. Felix Frueh is the president of R&D in personalized medicine at Medco Health Solutions, which is one of the largest pharmacy benefit managers in the United States. And Coin Hill is the CEO and co-founder of GNS Healthcare, a healthcare science and technology company applying next gen machine learning and simulation technologies to understand patient treatments. Again, we welcome your questions along the way through the dialogue box and through Twitter at #pmchat. So Eric, Felix, Colin, hope you can all hear me and welcome. Eric Schadt: Thank you, Kevin. Felix Frueh: Thank you. Kevin: All right. We don't want to spend a lot of time on introductions. But let's quickly go round the table, so to speak. And I invite you to introduce yourself, a little bit about your current venture or organization, and just in 60 seconds or something, just an opening statement, just outline the role that you are striving for your organization to play in the delivery of personalized medicine. Eric, let me start with you. Eric: Yes so the bulk of my time is spent here at Mount Sinai School of Medicine chairing the department of genetics and genomic sciences and directing this new institute for genomics and multi-scale biology. And my goal is to use personalized medicine to cure all diseases. That's the short answer. But seriously, what we're hoping to do here is figure out how to manage these large scales of data, integrate them with the clinical systems across the Mount Sinai Medical Center. They have everything on electronic medical records on EPIC, and to derive meaningful models regarding the diagnosis and treatment of disease, and actually push that information into the hands of physicians to impact decision making. So that's the charter and what we hope to accomplish over the next three to five years. Kevin: Excellent. Well, we'll come back and talk about that in more detail in the course of the next hour or so. Felix, let me bring you in. Welcome. Tell us a little bit about yourself and Medco. Felix: Thank you Kevin. So the Medco Research Institute is an entity within Medco as a PBM. And in particular, it's the entity that's dedicated to really utilizing Medco's assets on the big data side on the longitudinal pharmacy information at the medical claims data that we have from the millions of members for which Medco is providing pharmacy benefits too and really looking at what are the novel and cutting edge technologies out there including genomics, and proteomics and all these novel biomarkers tools to see what kind of difference that they can make in clinical practice. So what we're after is running clinical trials that measure, in particular, two things, the clinical impact they have and what it costs to achieve that clinical impact. Medco as a company has payers as their clients. And they're asking these two very questions. Does it work? And what does it cost? So the Medco Research Institute that I'm heading up

really focuses on how can we create value with these new tools and using the big data type approach with our databases that we have to create hypotheses but also using it as an infrastructure to deploy these novel technologies. Kevin: Thanks, Felix, just in case anyone is unclear on what a pharmacy benefits manager is, just in 10 seconds, can you just elaborate on what they do? Felix: Sure. In very simple terms, if you look at your healthcare card, it may say who your healthcare provider is. And then it also says who your pharmacy benefit manager is. Healthcare in the United States is split into the medical part and the pharmacy part. And Medco is covering the pharmacy side on behalf of payers. The clients that we have are health plans themselves, large self-insured employers, and then state and other government entities as well as Medicare. We cover about 65 million lives in the United States. So yeah, I'll leave it at that to make it short. Kevin: Thank you very much. And the third guest on our panel today is Colin Hill, the cofounder of GNS Healthcare. Colin, welcome in. Tell us a little bit about your background in your company. Colin, we're having trouble hearing you. I can't hear you. So you may be muted. Let's give that one more go. And if not, we'll go on to the question. Colin? Kevin: Colin, certainly I'm not hearing you clearly at all. So let's move on to our question, and maybe we'll work on that connection in the background. Maybe you'll dial back in. I'm going to ask Eric and Felix-- usually when we do these sort of panel discussions, it's sort of traditional here. You wait until the end, and then you sort of say, OK guys, let's look in your crystal ball and tell us what the future's going to look like. I kind of want to get that out of the way first. I want to assume that all of this stuff, the nitty-gritty that we're going to talk about today, the potential issues with regard to utilizing and applying next generation sequencing, and computational biology, and high performance computing, and big data analytics, and all these buzzwords, let's assume that all of that over the next few years really-- it just all clicks. It all falls into place. I'm just curious to know if we dare look 5 or 10 years ahead, what is your hope for personalized medicine? How are your organizations, your academic centers, your pharmacy benefits, your sequencing companies, how are they going to be driving and delivering this personalized medicine? What's it going to look like? Eric, let me start with you. Eric: Sure. So I think the 5 to 10 years time frame if this all works out as we all hope, it's going to be the very routine generation and use of information collected on individuals at the molecular level, so not just DNA, but other components, whether it's metabolite or RNA. And that that information is routinely used in assessing your overall health and well-being. So whether it's diagnosing risk of particular forms of types of diseases, or whether it's diagnosing a particular subtype in matching you to a treatment, that sort of information would be routinely used throughout the practice of medicine. Maybe 5 to 10 years is a little ambitious for that. But certainly at large medical centers where we can imagine that all newborns, everybody walking through the door has their genome sequenced, has other sort of traits collected. And that information is all crunched in the context of their medical histories and used to assess the overall state of that individual.

I think also we'll see emerging beyond the health field and maybe into other sort of less anticipated areas. So you could imagine maybe even dating sites, for example, starting to leverage that information to make sort of compatibility matches. I just see it sort of permeating nearly every area of our life as the generation of that information becomes very cheap routine, and people become comfortable with it. Kevin: And when you say, let's say the Facebook aspect, just one side from it, maybe some other people in the audience want to pursue you on that, but you're talking in 5 or 10 years about all this information, not simply screening for rare diseases, but you believe that this will make a genuine impact in terms of being able to predict one's risk for common diseases like diabetes, and obesity, and others in that genre. Yes? Eric: Yeah, particularly in the cancer arena. No question about it. I mean today in most cutting edge medical centers in the pathology department, they are already routinely starting to sequence in a targeted way certain genes to help elucidate for a given form save cancer, what subtype you might harbor, what treatment may be best for you, or how severe the progression may be. That sort of information is already starting to be generated and leveraged in the practice of medicine at these places. So I just see a broadening of the state of information as we go from the generation of the data to information, to knowledge, and ultimately to understanding. That information will become much more actionable and definitely affect clinical decision making on nearly every level. In 5 to 10 years, it may not be covering every single disease that we know about but certainly many types of cancer, certainly things like cardiovascular risk, or asthma, or other such diseases. We'll absolutely have models that inform on the particular subtype you have and what treatment might be best for you. Thanks Eric. Felix, come back in now. From Medco's point of view, you really are sort of at that interface with the patient. How do you see personalized medicine, assuming we can address all the many challenges and bottlenecks that we're going to spend the next 45 minutes talking about? But take us there, will you? Felix: Sure. So I'd like to expand on what Eric just was saying, because first of all, I agree with his vision. And second of all, I think some of the important considerations that currently are being debated are probably going to be going away. In particular, I think the issue of cost and access is one that is going to be solved by that time. I believe that the third component of quality with respect to whole genome sequencing will also be resolved. And what that means is that we probably all will have the genome as just yet another component at our disposal to make medical decisions. I think this is really going to be groundbreaking, because while we have been working on the last couple of decades on finding ways to get access to this information at a reasonable cost, we're now in the next 5 to 10 years are going to be able to see that. What we will have to be doing over that period of time though is making sure that we're creating an infrastructure, the logistics and the operations, that allow us to really harvest that information that's going to be truly big data in a efficient way. Because the amount that we can learn from that information, if we collectively look at the data rather than at individual genomes, just by themselves is absolutely tremendous. And I think if we are talking about finding associations with chronic and complex diseases like the ones that

Eric just mentioned, I'm going to throw cardiovascular disease, and diabetes, and these things into the mix as probably one of the most important considerations looking at what is going to cripple our health care system over that period of time most likely. The power of having thousands of genomes with their medical data and with their pharmacy data to look for associations and figuring out what kind of treatments actually work and for whom, you're going to be able to create an environment where a learning healthcare system is really what can become reality. And that's what I'm looking forward to. So if you're asking me what I would be dreaming of to have in 10 years, it is just that. It's a learning healthcare system where all this novel technology is part of the big data environment that we have access to. Kevin: Thank you Felix. Now I believe we have Colin with us. Colin Hill from GNS Healthcare, welcome. Tell us a little bit in brief about your company and again your thoughts as to how you hope personalized medicine will be medicine in the past 10 years or whatever. Go ahead. Colin Hill: Thanks Kevin. So GNS Healthcare started its life as a systems biology company attempting to fuse machine learning with simulation to really advanced drug development. And in the early days, personalized medicine meant for us the discovery of markers to stratify patients in clinical trials. We've really evolved into a big data analytics company where really the next generation of data crunching, where the data is not just genomics and clinical data, but even now involving longitudinal claims databases with clinical outcomes. But some of the data types that Felix has mentioned, this data which is the actionable data of actual daily healthcare, that this data is now becoming useful enough to use in figuring out what works in healthcare and for whom. It's the concept of personalized medicine, in my opinion, has really broadened and is starting to have more impact, not just in drug development, but in the application of care in the real world. Kevin: Thank you Colin and good to hear your voice. I want to follow up on that thought in just a minute because you brought up a term, systems biology. And both Eric and yourself have great expertise here in terms of mining huge data sets, albeit it's complementary sets. We'll go there in a minute. All three of you nicely laid out your vision for personalized medicine. Obviously, there's a big chasm between here in 2011 and where you all want to take us. And so I think we should do sort of sketch out for a minute where you see the biggest challenges. In framing this question, the list sort of goes on and on. It's where these roadblocks might be. It could be in the generation and the quality control of genomics and other research data. It could be establishing clinical validity and how we interpret genetics tests. The cost of all this is an issue. Navigating our healthcare system through insurance, and reimbursement, and regulation is a minefield. There's issues of data security and genomic privacy. There's the whole challenge of educating oppositions. I'm sure all of those are on your mind to some extent. I'll start with you again, Eric, and we'll work through the three of you. Where do you view, at least in your picture, some of the biggest challenges to implementing personalized medicine, Eric? Eric: I think the cost of generating the information and managing it in my mind not really such a big deal because those prices as the technology is evolved are just going to be driven down and down. And the systems to accommodate all that information will meet the demand. And so in my mind, it's really how do we, again, easy to generate data and

drive information from that data but very difficult to get to knowledge and understanding. So I think getting to the kind of level of understanding from that scale and type of information to actionable results that have actual clinical impact is going to be a major obstacle. Because with the common diseases in particular, it's not going to be any sort of constellation of genetic markers that initially is amazingly predictive for a given individual. It's going to have to be applying those kinds of models based on DNA information and other types of information to build initial models that you then iterate by considering the patient population is your trial population where you're iterating with physicians, and refining the models, and validating components until they get more and more predictive. That type of iteration is very, very complex. So I think how we get at away from just the research setting where understanding is being achieved and moving that into the clinical arena so that this is super effective is a major obstacle. And how you integrate that into somebody's medical record to the point where physicians can look at that information and without understanding all of the complexity understand how to act on that information for the benefit of patient, I think, is a major obstacle. So again, just to summarize, I think it's getting to knowledge and understanding in a clinically impactful way and then properly translating that into the hands of physicians and patients so that they can take the appropriate action are going to be, in my mind, at least where we're going to focus a tremendous amount of energy. Kevin: Thanks Eric. Let me go to Colin now and try to, at least in my mind, sort of do this in a somewhat chronological order in terms of how you guys are looking at the data. Colin, your business model has evolved a little bit as you were alluding to in your first response. But where are you seeing the biggest challenges? Do they have to be solved just with time? Do we need better models, different types of data? Give us your thoughts on that. Colin: Yes, I think the biggest challenge and opportunity is figuring out how to learn from observational healthcare data versus just attempting to learn everything from controlled randomized clinical trials. There is not the money to do all the clinical trials we'd want to do. But at the same time, we have $2.5 trillion of healthcare intervention that are being spent in this country every year. This is 17% of the GDP. How can we, one, collect this data in more real time? All the intervention, the stints, the drugs, the what have you that patients have received, how can we be set up to start to get this data in real time such that we have an always on clinical trial and then to use innovations and big data analytics to actually learn causality from these observational study? This in my mind is the biggest opportunity for us to figure out what works in healthcare, and for whom, and to be able to do that at scale. That is, I think, the biggest single opportunity in healthcare and the only way that we can improve health outcomes and reduce costs. Kevin: Thanks Colin. Felix, the question again is what do you see is the biggest roadblocks on route to implementing personalized medicine from Medco's perspective? Felix: Kevin, you mentioned quite a few aspects in your intro remarks to this particular question. I do believe that all of them are important. I do think, however, that perhaps the ones that are lesser mentioned or less well understood are actually the ones that

probably are the most critical. And those are access to the information and the operational and logistical infrastructure that is required to manage the data in such a way that it becomes accessible at the point at which it is actually important. To illustrate that, one of the often criticized issues about, in particular, pharmacogenetic tools is that while they exist, nobody is using them. And that literally is an issue of information education on one hand and access to the tool on the other hand. And in order to figure out whether we can create an environment in which that we can actually deploy these novel tools, we created commercial programs at Medco where when we see that somebody comes to the pharmacy with a prescription for a certain drug for which a pharmacogenetic test is available, we use that information in real time to go back to the physician and inform them about the availability of such a genetic test and what the test does, how it can be used to optimize patient therapy, and see whether or not physicians are actually willing to use that information then order to test and adjust therapy based on the information they've gotten. And just to give you numbers from two examples, Warfarin testing and Tamoxifen testing where the baseline is of uptake in the general population where you have an opportunity for a test is around 0.5% to 2% perhaps. If you look into our programs where we actively provide the outreach to physicians, you get that number up to about 25% and 30%. So this isn't just a percentage wise increase. This is a 20 to 30-fold increase in uptake. And this has something to do with the fact that you need to be able to create an environment and the operational infrastructure to be proactive. And I believe a lot of what we do still is in a reactive mode where we believe that building these tools is sufficient that somebody is then actually going to use them. But I think we have to be more in a push mode where we provide the information aggressively into the environment where it actually can be making a difference for patients. So I see this as a very critical part of what we have to work on. Kevin: Thanks Felix. Let's talk a bit now about big data analytics. And I'm going to segue this into our first question from Twitter. We use the term big data analytics in the subtitle to our discussion today. But I'd like to hear from the three panelists what they mean or what they understand from the term big data analytics. What are the big data sources that you're primarily addressing? Tell us a little bit about that and maybe how you hope to apply that to the delivery of medicine and better healthcare. So Eric, again, starting with you. Eric: Yeah, well, I think the big data analytics to me is how do you get your head around very large scales of data to able the human mind to recognize patterns in those data that may be useful for the diagnosis and treatment of disease and ultimately boil that down or fold those into the types of models we've been talking about to make predictions and classifications. So it's sort of not unlike what you would see quantitative traders, or do for predicting the markets in Wall Street, or for the credit card companies in predicting fraud. It's the same sort of idea. But in this case, we're trying to predict risk of disease or type of treatment one should get a hold of. And to effectively play that kind of game, you have to be able to represent and manage very large scales of data. You take a cancer, normal versus a tumor sample from a given patient and apply next gen sequencing technology to those samples. That's raw data wise. That's four terabytes of data, which in itself is a pretty big amount. But now imagine

that you're sequencing thousands of individuals over the course of a month running through your clinic. You're talking easily in the petabyte scales of raw data and on to exabytes is not at all a stretch. So how do you organize that information and process it in ways that capture the most information? I think we're being very naive at this point about what that taste. Because if you look at the cancer studies and that scale of data, what you see a lot of groups are boiling down all that amazing amount of data into coating snips, for example. So here you take terabytes of information on a given patient, and what you do is you boil it down to the hundred coding snips in genes that we think we know something about to explore how they might associate with that form of disease. And what we need to do is to start being able to derive way higher level information from all of that data without reducing the dimensionality to the most naive level and then being able to connect that up to other very large data sources. So the DNA dimension is one. But if you look at all the types of variables collected over a big medical center and that get embedded into the electronic medical record system, that's also a very big set of data that you want to connect up. And to connect those pieces up, you need the appropriate interfaces. You need the appropriate models for comparing those types of data and so on. So that's, in my mind, what I think about regarding big data analytics. Kevin: Thanks Eric. Colin, over to you. Colin: Well, I think in healthcare, big data is really coming from two ends. You heard from Eric the genomics driven end, which is certainly what makes the news a lot. But big data analytics also has a major place not just in the bio and genomics side of the equation, but ultimately, also on the paired provider data side where we're dealing now with large longitudinal claims databases, both drug claims and medical claims along with EMRs. These data sets are both wide and deep in that you're talking about instead of a clinical trial with a few hundred patients worth of data, you're talking about data sets with now millions or tens of millions of patients followed as a function of time. And so what this means is you can't use the same kind of standard tool in doing the analytics on these databases. This is not a matter of something you can do with standard statistics or using tools like that. One needs to be able to really do the analytics at scale in a different way that was done before. And that means not simply, of course, expert knowledge driven method. The haystack is too big for finding the needle. And ultimately, it's really automated method of doing statistical inference. That's what big data analytics means to me in healthcare. And there's a number of flavors that are starting, I think, to have some impact in real-world examples. Kevin: Thank you Colin. And we'll get you all to give us some sort of success stories or promising stories in a few minutes. Felix, the term big data analytics, is this something you're wrestling with? Felix: Sure, absolutely. And well, you wanted an example in a couple of minutes, but I think the best way to illustrate this actually is with an example on our end to highlight the way that we are going after using this type of information. I mentioned in my opening statement about Medco that we have access to about 65 million lives and longitudinal pharmacy data. In addition to that, we have medical claims data from about 12 or 15

million lives. So we can build correlations between pharmacy of drug use and clinical outcomes. And that is truly big data where these thousands and thousands of data points can be correlated to look for whether or not, for example, patients that are taking two drugs or more are at higher risk for experiencing either adverse events or poor outcomes over time. And one pertinent example is the use of clopidogrel, or Plavix, with a drug that interacts with the activation of Plavix. Plavix is a pro drug that only is clinically useful if you actually have the capability to turn it from being a pro drug into its active form. And drugs that block activation pathways such as proton-pump inhibitors may interfere with the clinical activity of Plavix. And therefore, patients who are taking both of these drugs may have poorer outcomes. So we look for whether or not, we're going to be able to see this in our database and realize that if you indeed taking both of these drugs at the same time, you have a 50% higher chance compared to patients who take only Plavix for coronary heart disease. So what's important with this is that we can observe this in a longitudinal fashion because we have that information in our database. And then you can do two things. First of all, we can cut down on the time that it takes for novel findings to be introduced into clinical practice very significantly. Because in essence, we can just pull the switch and say, well, this is important enough information that each and every pharmacist in our end needs to be informed about this so that they can provide this information back to physicians and patients and make sure that they're really only taking these two drugs if indeed they have to. That's number one. So we saw after we implemented this rule a significant drop up to about 1/3 in concomitant use of these two drugs, meaning that the proton-pump inhibitor has been dropped and that patients are really only taking Plavix or an additional drug that has the same effect as a proton-pump inhibitor. And then secondly, we can use this information for prospective clinical trials that can deal with genetics, for example. So in this particular case, that activation pathway goes through an enzyme that's a cytochrome p450. And we can now go out and genotype people to see whether or not if you have a certain genotype that may not allow you to activate Plavix. You also have poor outcomes. And with this, we're doing a step wise advancement of science in the sense that we're using big data to generate hypotheses that we can extrapolate into the clinic with novel tools and novel technologies to inform patients and physicians about what might be the best possible therapy. So now that you have a choice between Plavix and Effient and maybe soon to be other products as well, you can use these novel technologies to actually improve clinical practice. And I think that is one example where the use of big data in this step wise process has really cut down very significantly on time for introducing these new technologies into clinical practice. Kevin: Good to hear that it's already with us, at least a few examples. Let me take a question from Twitter. It's a good question. And anyone else who wants to follow suit can do so using the hashtag #pmchat, PM chat on Twitter. The question-- and I'll start with Eric again-- what computer languages and frameworks are going to be fundamental in carrying out big data analytics in a distributed manner? Excellent question. Eric? Eric: Yes, wow. I mean it's going to be pretty varied. I mean I can tell you the types of things we're exploring and using in that context is everything from R, to Python, to C

from our highly optimized routines that require heavy duty computation, to even Java. So the range of languages that are employed to construct analysis pipelines to carry out sophisticated algorithms on big data are very varied. And I think the bigger question is what is the right kind of platform that serves up those sorts of computer codes and hosting all of the data to enable the application of those algorithms to the data to persist the models in a way that people can interact with those models and to refine them over time as they're administered in a population similar to what both Felix and Colin have been talking about to leverage these patient populations as trial populations to really do all the refinements that are needed to get to the most accurate predictions? What we spend a lot of time thinking about are the compute infrastructures for carrying out the high-end modeling that gets done. Those are empty hard problems. They require very large-scale supercomputer resources to serve out. And then once you have those models built, you want everybody to have access to them. So how do you distribute those computations over big parallel compute resources? And how do you, on the fly, respond to user requests, whether it's a clinician or a researcher group such as Felix or Colin? How do you serve that information out to users as they request it in an accurate and reliable fashion? So it's all about, in my mind, getting to the right informatics compute infrastructure to host all of that information, and the modeling, and so on. And I don't know that there's any great solution for that yet, although we're certainly working with companies whether it's Amazon type services or leveraging the Facebook and Google type enterprises to build those types of platforms. But that's one of the major focuses of my group here at Mount Sinai. Kevin: Colin, do you want to add to that? Any specific computer languages or other types of IT expertise that you think will really serve people well in this field? Colin: Yeah, I agree with Eric for the most part in that it's not really so much about the specific languages. Certainly, things like Java and C++ are very popular for certain specific things in the field. But I think it is more about the distribution of the computation that's important given the scale. The term big data really came about first because the fact that data storage has gotten pushed to new limits. And so weve certainly been working with some of the big data database vendors such as Greenplum, which is now part of EMC, and others like Paradigm Four that really enabled this next generation of storage. But there's also not just the storage of the bottleneck, but ultimately the process power. And that means being able to use and leverage very large compute farms, whether they be cloud-based supercomputing through Amazon or some of the IBM Blue Gene machines, which we also use for some of the property. Ultimately, the problems that are special about this field come from the fact-- these are problems being done at a different scale, really an unprecedented scale compared to the past. And we're also dealing with patient level data. We can't forget that. We do have HIPAA, privacy laws, and other types of issues. So this means one has to also be able to do this in a secure environment. So it's a lot more stringent than the world of web 2.0 and the way that Silicon Valley typically operates.

Kevin: That's a nice segue into the first of what I hope will be a number of questions from the audience. I just noticed we actually got a string of questions buried on my desktop. So we're going to get into those. And if you have a question, do fire away. But at the moment, we're planning officially to end at about 2:05 Eastern Time. But we might be able to stretch it out a little bit. But, obviously, this a discussion that will go on for some time. We had a question from-- I recognize the name--Gitte Pedersen, the CEO of Genomic Expression. Hi Gitte. How do we get good clinical information on past health issues and treatments, she asks, when the average patient shifts insurance systems on average every 12 months and add to paper records? Felix, do you want to take that one? Felix: Well, that's an excellent question. I think it speaks to what we were discussing earlier with respect to the operations and the logistics that you need to ensure that this information, meaning genomic information in this case, I guess, or other large scale or more complex biomarker data is not just generated once and then, in essence, forgotten about. But it stays with the patient and can be used and reused again. I don't have an answer ready for how exactly this is going to work. What I do know is that we are diligently working on improving our IT infrastructure so that it accommodates much more complex large scale data, including whole genome information on the patient level. So I think it's going to be a effort that probably is going to happen at different levels in different ways depending on the institution. I know from hospital infrastructures with electronic medical records that some of them are considering putting that into their environment. So it's probably going to be a little bit of a hodgepodge over the next few years until we have perhaps more common data standards and ways that the data can be used in a more interactive fashion. I have no doubt that it will happen. But like many things in this environment, it's probably going to be multiple ways that will get us to where we want to go. And there will be a consolidation. Kevin: The second part of Gitte's question, the follow-up if you will, was how do we deal with security of sensitive genetic information when the value is in sharing our personal genetic information is not critical to ourselves, but equally for our children? Eric and Colin, you are experienced in putting data into the cloud and definitely in Eric's case, data belonging to large numbers of individuals. In terms of keeping this benefiting from the sharing, but keeping information secure regardless of what assurances GNS and the information act that gives us, so what are your thoughts in terms of securing and yet sharing data? Eric: Yeah, I mean on the one hand, the sharing is keeping things protected has worked out through appropriate DI identification of an individual's identity with the particular health information and in DNA say information, so it's not possible to identify the patient theoretically based on that information. And then in that sense, you're able to share, and I think the GWA studies have been really great there for demonstrating how to go from individual studies of 2,000 to 3,000 people on up into the hundreds of thousands because you're able to combine all that information into one large cohort and then carry out the analyses at that level.

So I think that it is possible to carry out that kind of research in this way. But then I would also say that the more open sharing of this information may just come about from new generations who grew up in this Facebook era and don't necessarily have the same kinds of issues regarding disclosing genetic information. I think what we're going to come to appreciate is as the technologies, whether it's IBM's DNA transistor or Pac Bio's smart sequencing in the next version that can generate entire genomes for not $1,000, but for under $100 or maybe even a dollar when it becomes so cheap that it's able to be sequenced out on the spot and at will, it's going to turn into like the video of your face. Ultimately, the courts will rule that you have no reasonable expectation of privacy. Because just like you have no expectation of privacy around somebody taking a picture, a video of your face in public, the same would be true with the DNA. It's going to have to be treated more in that manner because it'll be able to be generated so cheaply and so quickly. And I think even if we start thinking about any high dimensional data, many studies published, whether it's medical type information or your cell phone traffic, that any large-scale piece of information can be used to uniquely identify you. And we have studies showing just characteristics of your person, whether it's morphological features in your face, skin pigmentation, eye color, hair type. From that scale of information, you can start deriving genotypic bar codes to where it is then just like a video or a picture of your face. You have no ability to protect that information because you can infer genotype information from phenotypic characteristics that are strong enough. So I think ultimately, my belief is where this is just openly shared and not needed to be protected at all. Kevin: Colin, anything you'd like to add to that? Colin: That's ultimately where I'd like to see things go. I think there's more paranoia around the privacy of genetic information than is warranted. And people are suspicious of big corporation, certainly of health insurance companies. But from my experience in working with PBM and health insurance companies, these groups are very careful about making sure that personal information doesn't get leaked out, that that's not ever compromised. And so ultimately, the concern of people being disadvantaged in the health insurance context, one, it's illegal. And two, it's misplaced fear. And I think the upside from being able to combine large amounts of genetic data with clinical outcomes, what actually happens to the health of a person, has so much potential in terms of what we can learn that I do hope that we are moving towards that. And now these big blue buttons have been formed where individuals can just voluntarily- one, they can get ownership of their data, which is a big new development. And now that people can start to share their own data, this starts to provide an interesting almost self-organized way of data being generated. And I think it's going to be rather disruptive for the industry. Kevin: Thank you. Let me take another question from the audience. Folks are submitting them through the dialogue box. This is from Joel Dudley from Stanford, I believe, and involved in a start-up in the computational drug repositioning space. Joe asks, "Why aren't VCs funding more big data healthcare companies? Are there a lack of entrepreneurs in this space? Or are VCs having a hard time understanding the risk and value of these strategies? Or is federal regulations scaring them away?" Eric, Colin, I

just say you've all gone cap and hand to VCs from time to time. What's your feeling? Eric, let me start with you. Eric: Actually, I'd be interested to hear from Colin first. And I'll follow up with-Kevin: Colin, it's yours. Colin: I think the investment community is generally getting very excited about big data and big data analytics in general. And I think there's a growing number of VCs that are seeing healthcare as a very ripe space to place that in terms of big data analytics, helping to enable personalized medicine, helping to make healthcare more efficient. So I think that is starting. I did state the challenges VCs have seen are I think, one, they look at how difficult of a road it's been for some of the groups in the space. But I really think it comes down to business models. And when you get beyond the science, when you get beyond the clear market need, which the market needs super, super clear, and there's a number of VCs who I think really buy into that, I think it really gets down to business models. Who's actually going to pay? What are they going to pay for and when? And not sort of in the theoretical sense, but in terms of a transactional sense. Providers generally don't have a lot of money and are not always the biggest first mover buyers of innovation. I think it's a little bit different on the payer side. But I think ultimately, the challenges are large in drug discovery and development in terms of big data analytics because of the large cost, clinical trials. And I think on the payer provider side, there are challenges in terms of how it gets monetized. And that being said, I think you're going to actually see several companies get funded despite the caveats a given in terms of difficulty. I think the need is just ultimately too clear. But I think it is coming down to more business model and not just the science. I think the VC is buying to the science and the technology. That's a no-brainer. Eric: Yeah, and the only thing I'll add on top of that is really the actionability of the information. I think we in the US at least are somewhat impatient about the pace of progress. And just to remind everybody, it was 2001 that the first human genome was published at the cost of $3 billion in a short 10 years. After that, we're now sequencing for a few thousand dollars able to generate very large scales of data and are just now learning how to go from that information to knowledge and understanding. And so I think that pace has been very, very rapid, hard for VC communities really keep up with that pace and understand how it's going to impact. It's certainly when the information was first generated, there were questions about the actionability of what was coming out of the large scale GWAs and so on. But as we start nailing things down and getting to understanding, you're absolutely seeing a warming up in the VC community to these types of companies. And everybody sort of on the hunt now for the next Google or Facebook, which they think is going to be in this type of arena. So I think the future's very bright for this space. Kevin: Thank you Eric and Colin. A quick question for Felix. The finish line is ahead. We've got about 10 minutes, and then we'll wrap things up. So I won't get everyone's opinion on every question from now on until the end. Felix, this follows on from one of your earlier comments. Elliot wants to know, "Is there a clinical guideline indicating what

antiplatelet strategy gives the best outcomes in patients who cannot metabolize Plavix? And if not, what is the clinical utility of the test?" Could you talk me through about that? Felix: Sure. So there have been various pieces written by the American Society of Cardiology and other entities. I don't know if there is one official guideline that is out there given those new findings. But there are about 70% of the population that have the fully functional 2C19 gene. So they can perfectly activate Plavix. About 25% of the population has what we call intermediate metabolizer status. And then about 5% are poor metabolizers. And the poor metabolizers really are the ones that are most concerned because they may indeed benefit from using an alternative therapy. Effient is a competitive drug that came to the market about two years ago and is used as an alternative to Plavix now in certain cases. So I do believe that there are very sensible choices that now can be made. And obviously, the clinical utility for the genetic test is to evaluate whether or not patient may indeed benefit from therapy that is being prescribed. A big discussion is ongoing what to do with the intermediate metabolizers. There's studies that indicate that the tendency for them might be to group more with the extensive metabolizers so that they should be on Plavix. Some say that doubling the Plavix dose might be the appropriate thing to do. And others say they should go to an alternative treatment altogether. I believe that the next year or two will shed additional light on this. And there probably are going to be in all the guidelines issued in that context as well. Kevin: That's great Felix. Thank you. Let's go back to the big data issue. A good question from Ofir, "In what use cases does data or compute sharing between competitors make sense?" Eric, of course, you're involved with the Sage Bionetworks. Indeed, you left big pharma to launch this. So maybe you could just give a quick plug for what they're trying to do and talk about instances where it does make sense for competitors to share data. Eric: Yeah, I think the aim of Sage Bionetworks is to construct an open access platform of computing where data is shared. And models are dealt and shared as well. And I think the incentive for even pharma companies to share their data on that type of platform and have others combining that information to build better models, that they're going to get a better understanding of disease. So the incentive is going to be that if you achieve a better biological interpretation, a better understanding of disease, you're going to be able to place better bets early on in your development pipeline, whether it's designing a new drug or moving in to validate biomarkers. All of that can be dramatically increased if you boost your power by combining information and building more predictive models. The value then comes by a company's ability to iterate on that data internally to have hypotheses generate from those models, and then do experiments internally, and build confidence on one particular component. But the motivation is going to be you want the best interpretation. Your competitors are going to be getting the best interpretation by combining data and will have a leg up on you if try to hoard your data and analyze it in a silo. Kevin: Colin, anything else you want to add to the data sharing question?

Colin: I think Eric covered it pretty well. I would just say it's really critical as for the path forward. One can't learn mechanisms, drivers markers, from too small of a data set. And the only way sometimes to get sufficient numbers is to find ways to share and combine. Kevin: That's great. Another question from the audience, "Other than the people on this panel, who are the big players in big data analytics on both the genomics molecular biology side and the healthcare side?" Who would like to take that first? Eric: Well, I mean I could say I think there a large number of groups, I mean certainly the Erode, one of the biggest data generators on the planet as far as DNA based information goes. You have BGI in China as well as, just an amazing force today to be reckoned as far as their ability to generate very large scales of data and derive new meaning from that data. Any of the genome centers, whether it's WashU, Baylor, all have at least the ability to generate big data and then represent that information. But downstream data analytics, integrating that information with other types of data, I think the list is much smaller. I think many of the key players are on the call. I think groups like Andrea Califano at Columbia are good at this game. They're groups at Stanford, Daphne Koller, Gary Nolan, those types are doing some pretty interesting stuff. So at this point, I think handful of groups that are able to do that. And then the health care sector, I think Felix probably in the best position to answer that. Felix: Yeah, I do think that you have to distinguish between the ones that actually are generating the data on the healthcare side and then the ones that are sort of overlaying technology to extract the knowledge and information from such data. Certainly, healthcare providers, and PBMs, and the Kaiser's, and these groups, they have large data sets. And one can only fantasize about having the additional genetic and other biomarker information overlaid with their clinical and pharmacy data. So I think that's where the data will accumulate. And then we'll have to see what kind of novel technologies and tools that we can use to actually extract information from it. I think these are really the two separate aspects that we need to consider. Kevin: Thank you. Let me just try to squeeze in a couple of final questions. And then we'll get final thoughts on the panel and wrap things up for today. A question from Trish about ontologies and natural language processing, "Are you using ontologies and natural language processing to identifying knowledge in text and aggregate this information rather only using codes, for example, such as ICD codes?" Eric: Yeah, I think in our case, the answer is absolutely yes. In fact, it's one of the sort of newer areas we've been exploring and integrating all the molecular data with clinical data using Bayesian Network Frameworks. We are now actively in sort of an natural language processing way, deriving information from records as well as literature to come up with prior-- we're effectively using it as prior information on building the models and building the networks. So that's sort of where the work is at today. But I think definitely a tremendous amount of energy and activity behind that, I think it's going to be sort of ripe for how to do that most effectively to combine with all the other sort of biochemical molecular data.

Kevin: Anyone else on that one? I'll take that as a no. Felix, this question which we got by email about monetization, "How are you seeing physicians paying for these new services? How do they see big data being monetized?" That's a good question. Felix: So I'm not sure whether we're talking about big data in this particular context. I think that the access to novel technologies and the use of what actually comes out of analytics of big data to treat individual patients is where we have to find out how we're going to be able to pay for it. And in the cases that I mentioned before where we have commercial programs for certain pharmacogenetic tests today, those are covered by the payer. So our clients, the payers, they're signing up before they know they're enrolling in these programs, and to test this is fully covered, and the services that go with it as well. I think that the use of big data is going to be dealt with in a different way where companies like Colin's, or also Pac Bio with Kevin, and other players in the field, they're going to be making their money on the front end by actually providing information and knowledge about how to use this information. And then the translation into clinical practice is where it's going to be applied on a patient level basis. And that's where we have to find a solution on the reimbursement side. Kevin: OK Felix, thank you. We're going to close in just a couple of minutes. I'm going to get the final thoughts from Eric, Colin, and Felix. I note before we go that the slides and the recording of today's webinar will be made available, and sent out to everyone on the call today, and maybe posted online. And if people have questions, they're welcome to email the organizers at the following email address, personalizedmedicine@scratchmm.com. I just want to get your final thoughts. I promised that we talk about success stories. And so if you want to devote your final sort of 30 seconds, your final summation to an optimistic note, a success story perhaps, you're welcome to do so. If you want to spotlight some of the things we need to really focus on to get right, you're welcome to do that as well. Eric, let me start with you. Eric: Yeah, well, I'll just say one of the more exciting things at Mount Sinai has been we were awarded one of eight or nine of the big grants for the Merge study, which is can we leverage EMR based data to actually have a clinically actionable impact on outcomes for cardiovascular disease? So I think that sort of active trial's going on in real time, leverage all the EMR type information and big medical center settings where that information is going to be shared and combined across nine different major medical center sites to really provide more definitive evidence of how this type of information. Everything we've been talking about absolutely will lead to novel discovery that's clinically actionable. So I think the future's very, very bright. And I just want to thank everybody for their attention on the phone. Kevin: Thank you very much Eric. Colin? Colin: What I'm really excited about is the fact that I think the convergence of both technology and science, as well as just business need, the crushing economics that we're dealing with in healthcare, are converging to make this new world possible. And there's a couple of specific things that I feel pretty happy about that we've done recently. One is in collaboration with Biogenetic, which is where we made really the first data

driven personalized patient model of rheumatoid arthritis drug response that involved anti-TNF drugs. And I think, while this is not the be-all and end-all, to me, it's now possible to integrate multiple layers of data and get at true patient level predictions where you can take blood from a patient and actually get a prediction as to how they're going to respond to Remicade, Enbrel, Humira, or any of these other drugs, And then on the sort of the observational data side of the house. I'm actually, recently working with Felix and his group at Medco, really just seeing for the first time a true hypothesis-free way of approaching longitudinal claims data. And these are two examples of a lot of new things to come. And I'm pretty bullish on the future here. It is starting to really work. And Felix, you get the final response. Felix: Yeah I want to give one example where I'm really excited about looking at additional and perhaps more unconventional uses of genetic information. It's in the area of adherence. You know that drugs usually only work when you actually take them. And if you look at how dismal adherence statistics are, you realize that, for example, in the area of statins half the patients don't take their drug anymore after about six months. So what we were interested in is whether or not providing genetic information about risk can encourage patients to be more adherent to their medication. And we've recently completed a study in collaboration with Pollara on this particular question. And while we haven't published the results yet, I just want you know stay tuned because they're going to be published very soon. And they're going to be very exciting. So I believe that we're starting to scratch the surface on what the use of the human genome actually can accomplish. And they're going to be very many additional uses. Eric, in the beginning, alluded to that as well, that we'll see additional benefit that perhaps at the moment we haven't necessarily been considering. So this combined with access to large databases, and their appropriate use, and the protection of privacy with it, I think is paramount for the next few years. And that's why I'm excited to do what I'm doing and look forward to be part of this personalized medicine movement to come. Kevin: Well, thank you very much, Felix. And thanks to everybody participating on today's webcast. We've been discussing the new face of personalized medicine. We've only just scratched the surface. And I apologize if we couldn't get to anybody who sent in a question. We were gratified by the number of questions. And they were all very good questions. And I hope we can convene a similar forum in the not too distant future and take the conversation a bit further. And so on behalf of Eric Schadt from Pacific Biosciences and the Mount Sinai School of Medicine, Colin Hill from GNS Healthcare and all of his colleagues who helped put today's webinar together, and Felix Frueh from the Medco Research Institute, I am Kevin Davies, editor-in-chief of Bio-IT World and author of the new book, The $1,000 Genome. It's been my pleasure. I hope you've enjoyed it. And thanks to everybody. We have more than 100 people on today's call. And I hope you were able to derive some pleasure and some useful information from it. We hope to see you and hear from you again soon. For now, signing off. Kevin Davies. Thanks for listening. Bye bye.