 Well, thank you very much for the opportunity to come to this meeting. And what I try to do in a couple of slides is just summarize the sorts of thoughts that I had as I listened to the discussion over the last day. And as I mentioned earlier, in doing that, I tried to put it in the context of kind of reflections on what I had observed at least in terms of previous experience with GWAS in complex diseases. And there are these different kind of pools in different directions between the many smaller studies versus a few larger experiments. And I think that one of the lessons from GWAS, as I think I mentioned earlier, was that, and as people are reflecting, that probably we need to be moving towards few larger experiments. As was illustrated by the need for the meta-annouses of the smaller studies of GWAS. So I think there's a very clear agreement on that. There's been a lot of discussion in the past around the kind of deep versus, no one's used the word shallow, but I assume that's the opposite of deep, participant phenotyping. And yet, again, as I mentioned during the last day, there was all of this focus in many of the different GWAS, indeed family-based studies, some of which I was involved in, where we spent days, weeks, months arguing about what was the definition of the disease, what were the kind of characteristics we were going to record in the participants. And then we've seen with the meta-analysis of all sorts of different kinds of studies in GWAS that that has been largely ignored or found not to be particularly important in terms of identifying variants that were associated with disease. So I think that it does argue for, perhaps not shallow, but for these large cohorts that not having very intensive phenotyping does make sense as a strategy. The point that perhaps I hadn't heard brought out during the discussion and which goes to the title of the meeting, which is the difference between cohort studies and large sample collections, is the disadvantages or advantages of the different approaches. And if we look at the GWAS studies, they have tended to be focusing on a disease and finding variants associated with a particular disease. Whereas a prospective cohort doesn't focus on a particular disease, it allows us to look at the associations of risk factors with lots of different diseases. And so if we have a few large experiments, a few large cohorts with relatively shallow phenotyping, then we're able to, within those few studies, study the genetic and other determinants of many different diseases. And so the longer it goes on, the more valuable it becomes, the more return we get for that investment, provided that we can link those cohorts with health records and ideally with extensive health records. Again, if we talk about cohorts, one of the big failings of prospective cohorts is that they tend not only not if they're big to phenotype participants, but even more so they tend not to identify or indeed phenotype the diseases terribly well. The criticism that's often made about prospective cohorts, that if they involve linkage, then they link to your death records or to cancer records. But information is not obtained about those diagnoses. And so I think that there is a question about how deeply to phenotype the disease. And we probably need to do a bit more in terms of phenotyping of the disease in these cohorts than has been done typically in the past. And that may be another example of identifying people with disease and then going to those records and getting more exquisite phenotyping of the diseases. So it's about the disease rather than the participants. And again, I think if you look at the GWAS experience, what we've seen is that it's become international. And that probably if we're thinking ahead that we should instead of having got to this point retrospectively with GWAS, that we ought to be planning now for it to be international, not just have it occur through necessity after 10 years of studies that haven't produced results and being kind of forced to work together, that maybe this is the time, as with the Neanderthals, to share our DNA. And in that respect, I also think that it's important to not have a short-term agenda driving the strategy. And I think Peter Donnelly raised this point about if we do XM sequencing in a few years time, we'll wish we hadn't done it because we'll throw away the data and have whole genome sequencing. So maybe we should be thinking not what do I want in three years time, but what in 10 to 15 years time will I wish I had done. And I think if we turn it around that way and think about what should we do in the next few years that we will have wanted to have done in 10 to 15 years time. And that may be much more about bioinformatics and analysis of data than it is about actually understanding the genetics of disease. That maybe we should be not allowing the constraints of the cost to drive the next few years but think about it as an opportunity to do relatively small-scale studies where we learn how to use the data, how to make sense of the data so that when we can do it on the right scale, we then do it on the right scale and we know how to analyse those data rather than wait until we've got the data before working out how to analyse it. And I just wanted to touch on this thing about disease adjudication because I think the sorts of things that we see with eMERGE and other kinds of settings where we can put cohorts or establish cohorts or have cohorts within a health record linkage system really does allow us over the long term to build a resource that will really allow us to get enormous information about the genetics and other determinants of disease. So I think that if we're going to have some large scale studies, we need to have very scalable approaches to identifying disease. And I think the other important thing is we need scalable approaches that can identify a lot of different kinds of disease. So rather than having the focus beyond diseases that are well studied because we're able to find them like death and cancer, we want to be in a situation where as with eMERGE and these other studies, you can link to other kinds of information, hospital discharge records, primary care records where we can perhaps use new technology to find out about health outcomes easily on a large scale that are not picked up from record systems. So can we use e-mail to assess people? Can we use e-mail to assess their cognitive function and cognitive decline over time? Or can we use it to detect other things that are not detected well by primary care or hospital, depression, musculoskeletal problems, the kinds of things that perhaps haven't been studied well in these kinds of studies? And then to use those crude systems to identify cases and then use more specialized record systems. So I'm still not talking about getting back to the participants all the time, but using other kinds of more detailed record systems to find out more about the individual. So there might be particular registers of cancer or of morbidity in Britain, for example myocard infarction. Can we cross reference different electronic record systems to adjudicate or confirm that a person has a particular disease? Could we send out kits where people would have a blood sample taken and that would help us to confirm the disease? And then go even further to build up a resource in terms of the phenotyping of the disease. I think Julie described what's been done in the health professionals study or nurses study around getting tumor collections where people have been diagnosed with cancer so that the cancer itself could be analyzed. Can we get imaging data about particular types of outcomes, say stroke, in order to determine what kind of stroke and to subclassify it in more detail? It would be remiss of me as the PI Biobank not to describe it, but it has a number of these characteristics. That's not to say there aren't other cohorts like that in the US that clearly are. But I think that it does reflect the kinds of things that have been talked about today and yesterday around having large prospective cohorts, the ability to look at a particular exposure on a range of different diseases or many different kinds of exposure and how they interact with each other, to have sufficiently detailed information about a large number of people and then the potential to enhance the phenotyping in large subsets, for example the proposal that we have to image in 100,000. And again going to the points that were made around access, the ability for researchers from anywhere in the world to access the resource, to use it for any kind of health related research, to have the ability for recontact. I think these are all the kinds of things that we need to be building into at least a few large cohorts that may already exist within cohorts in other parts of the world and in the US, so that internationally we have the potential to do these large sequencing studies. Those were the kind of thoughts I had as I listened to the discussion. It kind of feels as if we should perhaps not do too much over the next few years and just wait a little bit before you spend a huge amount of money on sequencing, just wait until the cost comes down. I think there's a danger of going too fast actually. So I would strongly endorse the idea that we need to think of this as a project that's going to go on for generations, but 10 or 15 years is a good timescale. It's roughly the one that we chose for the HGP, what we want to have accomplished at the end of it. On the issue of going too fast, that is a risk. The HGP approach was really borrowed very directly from Fred Sanger. If you look at the history of Sanger's immense contributions to DNA sequencing, he tackled projects which scaled up by about a factor of three, and projects where he could bring a key phrase in the Albert's Committee report about the human genome project was pilot projects of increasing scope and scale. I think that's exactly what we need here. We just want those projects to have the characteristics that they really are building toward this different discovery model, as I've called it. I think that was the point I was trying to make, that we designed the strategy for the next few years based on what we want to have done in 10 to 15 years rather than having a kind of interim strategy that doesn't build towards that. To that end, I fully agree that we need to have pilots and we learn as we go along. If we just take your analogy of the GWAS, we started with 500 or 1,000 cases and people said, well, let's wait until the chips are cheaper, and people just kept pushing along. I think we learned a tremendous amount such as the point that when the chips got very cheap and we had enough people scan, we had worked out an awful lot about the meta-analyses and imputation, so the discovery was continuing with the addition of these new technologies. I have a little bit of a different view. I think we should design very good pilots that go as fast as we can knowing that we're going to revisit these things and that for the numbers that we've talked about, we're going to have to get very large. I think whatever pilot we come up with is going to be inadequately powered to adequately address many of the things that have been discussed around this table today. It's going to be in the McDonald's supersizing of this that may not be just this project but things going on in the Netherlands and Estonia and Spain, England, China, lots of other projects that will hopefully have what I think you pointed out so very nicely, that the international collaborative spirit, this group isn't going to do it all by itself. I would like to just emphasize the value of doing as much in those pilots as we practically can knowing that we'll make some mistakes and we'll just get better as we iteratively go through the data. I guess what I was trying to say was that having learned from GWAS, it might be an idea to work together prospectively rather than work together retrospectively and that one could avoid a lot of unnecessary replication by doing so. That was really the point. But I think we're at that point. I mean that's the beauty of GWAS. It's set that standard. I think many of the people around this table and who are part of the studies touched the people here have been in meta-analysis and it's not an uncommon thing. There are many primary data sets that are showing up for the first time as part of meta-analysis as opposed to the old 14th century. I need to put my flag in the ground and have my paper that shows that my study is out there. We've seen that sociologic transition already begin. That wasn't the 14th century, that was about 2009. I know. Well, but if you take genetics into the year zero AD, that would probably correlate a niacriffer. The only comment I'll add is in some ways ESP charts the type 2 diabetes consortium. We have many things going on that could be considered pilots that we can learn on from today and I'm sure in the cancer area there's many things going on. Paul? Yeah, Paul Sorley, Epidemiology branch, NHLBI. Just to step back a little bit, I think, since this is a wrap-up, there's an awful lot of people out there and maybe in this room who think that all the money that's gone into genetics has essentially gone into a hole without a bottom, probably, and that in the larger scheme of things for public health the advances from genetics have been really very, very small. If you're going to put lots of money into another technology, and they will say another technology that's really cool, I think you're going to have to do a lot to convince people that you really are going to do something big here. Two statements were made and they were sort of reiterated here. One was in terms earlier in terms of goals. They need to be big goals. I think that was the word. You've got to be substantial to say to people that you can address the causes and prevention of disease. The second is related to the pilot. You've got to convince people that, in fact, this is feasible, that it's more than just a new technology, but you can really make some advances in this. Otherwise, I think it's going to be pretty hard to convince some people to put a whole bunch more money into this. Public health. Exactly. I agree with you, but I think we have to also take stock of just putting the genetics and genomics in the context of discovery if we think of other things, the history of cardiovascular surgery or the history of cancer therapies. Many of these things date back 40, 50 years ago when they were 10, 15, 20 years of very difficult pilot studies. NIH has a whole blue building that was built for doing cardiovascular surgery, and the number of difficulties and things that they learn were the basis from which then cardiovascular surgery really took off in the same way with cancer therapies. We have to be very careful, but we also have a history that we see behind us. From that, I think we have to argue on the basis of having very important goals, but knowing that this is a long-term project and that we don't get the quick fixes, and that's where the GWAS age really did us in with people declaring that we would be able to have these tests and explain the heritability of all of these complex diseases in a matter of two or three years in a sense that we did ourselves in on that. In thinking about this next pilot or this next set of projects, we have to be a little more realistic in that larger context. I just want to make a comment about Maynard's statement about IT, which I guess I'll exaggerate a little about, don't freak out, and citing credit card systems, bank systems, streaming as evidence that big problems can be solved, and I'll just make the point that the people who did that or the powers behind it invested huge amounts of money, and that in biomedical research, we tend to be very cherry with the amount of money that we want to spend on information systems, and that's something that we need to really change. Chris, just to follow up on Paul's comment and also a little on Rory's kind of triangular caseness being important, definitely agree that if we're going to be ambitious, we're going to need to make some, set some really tight priorities, and one way to think about this is diseases, specific, rather than having a thousand flowers, maybe a couple of diseases, within which there could be, if you have control population, like a case cohort, you could be looking at a number of other things within that control population, so that's just one thing to consider, and then the other is in terms of, I think we need to push both the phenotypers and the sequencers. We've heard this yesterday, I guess, and I think it was $10 million for 2,000 whole genomes, but the cost issue there is the coverage. So if you have lower-fold coverage, you could potentially sequence many more people, but we didn't hear any discussion about that. Similarly, I brought up the comment about the exome chip and the metabolite chip, et cetera. Maybe that's a way to make the dollar go a little further in terms of much larger cohorts. So these are the kinds of things I think, but on the phenotyping side, it's the resolution of the phenotype. Maybe the UK Biobank, you could, before you get into the sequencing, select 5,000, 10,000 of case X and do that detailed phenotyping and say these are bona fide cases. They're not just kind of like hospital discharge diagnoses or similarly in a merge. So it seems like on both sides, and this is going to require cooperation, no doubt, both within the US and internationally. Rory, I wanted to make this or have the same question as I asked Dan earlier. One of the things I'm concerned about if we require, we link this too closely to this word on electronic medical record. In this country we don't have, we don't have a national health service, we don't have an electronic medical record. And I want to make sure that in designing a study moving forward that we don't wrap it too closely around an electronic medical record which would exclude a huge proportion of this population that's already under-stirved, and this would just go further to make them more underserved. Well, I mean, I absolutely agree with you about the value of heterogeneity. But you don't necessarily have to have all of the heterogeneity in all of the studies. I mean, you can have, if you have a few major studies, then you want to have a few studies ideally in different populations that increase your heterogeneity. If you want to study Asians, go to Asia, don't take them as a minority group in the US. So I agree, you want heterogeneity and you can do that between, by having studies in different populations. And I think that's true of socio-economic aspects which you raise. Where you can get socio-economic diversity within a cohort, then get it there. There may be other kinds of diversity that you can get in a study in the US. But you don't have to have everything everywhere. Maybe I could also comment on the fact that sadly in the US, people who don't have access to medical care or medical insurance or electro-medical records are not that unusual. In other countries, there are really strange people. In this country, they're not. So we're probably capturing a lot of the characteristics of those folks by getting a broad range of people in the US, even including some of those who do have access to medical care. I was just thinking, you know, I think we did a great job at the end of talking about our criteria for selecting samples. But I was trying to tie this together with our first slide today, which was on sort of the scientific questions addressable by sequencing. And I was wondering if we ever talked about the order, meaning that were we thinking we would pick sort of a couple that were just the ones to go with. And then while you're putting together the samples to sequence, it's sort of making sure that they will cover what's needed to be able to do it. Or is it putting together a resource and then of all the wonderful things that could be done with it, we'll pick afterward. And it makes a difference just sort of logistically. Because if, for example, we decide no matter what we do, we better make sure that diversity, by ethnic group, we really ought to be able to make comparisons. Then the way that you put together this sample will be very different than if we just put one together that represents everybody and then we play with it. And I'm not sure we did that too much. I couldn't agree with you more. And I was actually thinking a few minutes ago, but almost exactly that. I think actually that's going to be Terry and I's biggest challenge in putting the document together from this meeting. It's frankly, today it was hard to do everything in a single day. But I think you're absolutely right. We have to have these recommendations from the afternoon tie back to those use case and errors or questions from the morning. Right now they're not, but I think we can do that. Go ahead, Nancy. Okay.