 So Stephen and I put together a few, identified a few themes, general themes for discussion and prioritization, one around translation, the other one around genetic architecture, nested in both the study design and the depth of the various human tapping in the various projects, but that's really an issue of the next session. But for translation, the discussion centered around using translational paradigms, discoveries for either advocacy, adverse reactions, studies, or risk resilience outcomes, some of these studies could also be used using modifiers, some of the translational studies could be used with modifiers, maybe perhaps Mendelian, some of the Mendelian disorders and Max beautiful talk on CF, oligogenic diseases or even complex diseases like autism or others. Again, it will depend on the study design, which one of these translational hypotheses we could pursue. The genetic architecture theme centered around discussion of rear variation, clearly we need large sample sizes. I think there was consensus. And we also talked about the need to develop that the mathematical framework of these tests for variation is still underdeveloped. And we also touched on higher what we call your higher order models. We think about, again, modifiers, epistasis, and other higher dimensional data that could be integrated to the point of your systems biology perspective, as we heard from Julia. And then lastly, and following that great discussion we just had on phenotypes, how important are phenotypes? Our study designs were to broad, in-depth phenotyping, and so on. And with that, I want to open it up for discussion. What did we miss? Yeah, what did we miss? That it's burning. There were a number of comments. I don't think there's enough about very hard to see the spectrum of phenotypes with a given mutation. So we're trying to get questions that we want to address by sequencing. So that isn't a sequencing question. It's a phenotyping question. And yet you have to do the sequencing to find the people that have the variant to then go in and phenotype them. So is that kind of covered in what you've shown? Well, I think that's going to be this afternoon. And it's been referred to last night. And today is with this question of revisitation or recontact with the subjects. And I think it maps also back to what Maynard was saying about being able to sequence as large as possible and finding people who are under those thresholds of current clinical diagnostic categories of saying they have IBD or whatever. And as you sort of look from the genetics and try and work backwards to what you think they're interesting phenotypes that are lining up, it's almost like reverse GWAS, so to speak, is what I'm hearing, where the genetics are going to drive how you would group your phenotypes in what may need to be done to refine that. There was an entertaining discussion that went on for about 10 years about what was forward genetics and what was reverse genetics. This is the true reverse genetics. Genetics traditionally went from phenotype to genotype. But I think this discovery by genotypes is going to be very important. Now, let's make one more point on this issue of phenotypic heterogeneity and quality of phenotyping and so forth. I think there are two issues here, probably more, but there are two really obvious ones. One has to do with whether the phenotyping was simply done in the sort of the best possible way. So I don't know anything about measuring heart rates, but I'm willing to assume that there's some better way of doing it than just sort of trying to count your pulse. And so there, of course, we would like the phenotype that we're working with to have been at least measured in the best possible way. Usually when there's these discussions about phenotypic quality, don't so much have to do with that. They really have to do with just sort of how precise and uniform the criteria were for establishing very complex phenotypes. Tom might want to comment on this because he's a veteran of the sort of mental health wars. But as an outsider observing that, it's an interesting example. There was a 1980s phase of false positive linkages to this and that psychiatric phenotype and a lot of hand wringing about that. And it tended to lead to a long period in which the idea was that we'd find the genetic causation of these diseases if we used really strict uniform cross-center kinds of criteria for exceedingly complex traits that we don't actually really know how to phenotype. I always thought that was mistaken, well intended as it was, according to a very simple argument. And that is that if you look in the literature, I believe this comment applies rather broadly to many human phenotypes. But if you looked in the literature about psychiatric phenotypes and said, OK, I want to find the patients where I have the strongest belief that I'm looking at genetic causation, what you would, of course, do is find these very rare families, but some of them have been published on repeatedly over decades, in which in multi-generations, you just look at it and it looks sort of autosomal dominant for psychiatric disease. But the phenotypes in those families are all over the mat. You've got schizophrenia, you have paranoia, you have bipolar. This is not because they were evaluated by different psychiatrists. Because their phenotypes are all over the mat. They're being modulated by a lot of genetic and environmental factors that just lead to all sorts of different phenotypes, even though there probably is underlying all of that segregation of some mutation that is perhaps best thought of as one that is affecting the homeostasis of normal behavioral brain function. And I think there's probably lots of that. We just tend not to capture it. And so it's a more biological way of looking at this issue. Yes, we would like to measure the phenotypes with the best kind of thermometer and so forth, but I think that we want these cohorts to have been minimally pre-selected to meet a whole bunch of criteria. Could I ask one question? If I'm understanding correctly, maybe I'm confused. You're suggesting that we'll do genome sequencing, we'll find variants, and it will be associated with the disease, recognized disease. And some people, it'll be not associated with recognized disease in other people. So we're going to bring those people back in and phenotype and see if we can pull out some phenotype that's not been recognized. But isn't it just as likely or more likely that there are underlying genetic variation in those people that don't have a recognized phenotype that's modifying the situation? And so before you bring the people back in and phenotype them, how are you going to look at your genetic data to figure out if they don't have a clinical phenotype that's been recognized because they don't have a clinical phenotype because they've got other modifying genetic or environmental influences? I mean, that is the $64,000 question. I mean, it takes us from the single gene, and now we're down to single variant testing. And then the question is, what's that context? CF was a monogenic disorder, but everyone forgot about the 24,999 other genes. And now as we have the tools, we can see that that has an important influence. I think we're going to have to think those same things through. And I wonder whether there'll be sort of a dialectic between how good the assessment of some of the phenotypes are and the capacity to be able to look at ancillary or related things, and the genetics just not of that variant or that gene, but what the context is. I mean, because in the end, we probably all have our own genetic carburetors set at slightly different levels, but we have no idea how to quantify or to describe that. We just sort of, I think, generically think that everybody has a carburetor, but the question is, what's it set at for me or for you? And it's going to take very, very large numbers to do that. And that could very well be the most important thing that comes out of sequencing very large numbers just to see the relationships between these variants and what they may or may not mean in very broad strokes. I wonder if it could work like this, because we've said bring people back, but sometimes you don't need to. So if Judy's work in an exquisitely well-characterized population says, oh, interesting, not just for CF, but I find something which leads to COPD in that population and now let's say it wasn't known, I want to just bring it back to this central, cheaper, broader, available, large resource and quickly look to see anybody else that we've already sequenced who has anything remotely like it or otherwise genotype. Then I don't want to bring them back in. I just want to look, for example, at whether or not they ever were hospitalized for COPD, which we can do now. That doesn't cost anything. That's just linkage. Nobody's been brought back in, or if it's in CRUK, maybe they were looked, or not maybe they're in a, it depends on what's sitting you're in, but you do a quick look and yes or no, there's a signal to this more common other phenotype. And if you like what you see, it's anything, it's not perfect, then step three is you go do the detailed appropriate study there. You can't ask this broad, smart, quick check place either to be as clever at discovering what Judy found nor could you ask it to be ready made to answer the question on some link to the other disease. So I feel that this kind of three step bounce or three different places to go might be closer to answering the question we're trying, the design question we're getting at here. We want something that can fulfill that middle function really well to save us a lot of time chasing false leads. We'll miss a few things, but it should be at least the place to go for her to come and say, I know it's true for CF people, let me look at some others. If I get any kind of hint, then I'll spend the money and do the more elaborate study. But you don't have to always bring people back in. Sometimes you can see, what did they die from and with what were they hospitalized? And on that note, you can also get a lot of phenotype data just by asking people. So simply by having a re-contact mechanism where you can simply ask questions of someone who could potentially be very powerful for certain phenotype. Would do that, would most ask by questionnaire. I mean, you take nurses or something, it would be the next mailing, it would be unlikely that they would bring them back in and draw blood and do a whole series of things. Just getting back to the therapeutic development question and Judy's talk, I was really intrigued by the TNF story and the fact that there is although this is biologically a very compelling, very clearly established pathway in this disease, there's no genetic evidence to support, well, weak genetic evidence to support its role. And I was wondering what you thought was the explanation for that and whether it teaches us anything about the challenges for using genetic information to find therapeutic targets. So I wouldn't say it's no evidence because literally one of the major downstream pathways of TNF is NFCAPB activation. So we have a number of NFCAPB genes that are associated. The other kind of point that I rushed through was that on our go analysis, again, we haven't proven that these are actually the causal genes, but in our go analysis of the genes within our loci, we see a large number of ubiquitinating and deubiquitinating genes. And so a lot of how TNF works, as well as immune-mediated diseases in general, is you have transient effects in innate immune cells. And so you have this rapid increase and decrease of RNA expression. And so one of the most important pan-immune-mediated disease genes is A20 or TNFAIP3, which is like TNF has all these A-rich elements and goes up and down. So the fact that we have 15 ubiquitinating and potentially 15, we haven't proven them, but ubiquitinating and deubiquitinating genes, that's a lot of how TNF acts. And so the answer is it's just much more complex. The IL-23 pathway is so simple and straightforward. TNF is almost by itself too important, I think. I mean, to think that blocking one side of kind would work in a disease, I would never have thought of what's gonna work. The fact that it does is so remarkable. And so I think a lot of the other mechanisms that are being utilized now for drug discovery involve much broader suppression of cytokines than just blocking one cytokine. So the fact that blocking one cytokine can treat diseases, I would say remarkable. So it's complex as the bottom line. But it certainly emphasizes the fact that a lot of the genetic evidence for drug targets may well be indirect. So we may need to take the preponderance of genetic evidence and use that to point back towards some therapeutic target, if that makes sense. Yes, again, the flip side of studying the cytokine story is that the clearest case of them in dealing IBD is IL-10 deficiency. The knockout mice gets IBD, autosomal recessives in the cytokine receptor, and yet IL-10 has not been effective in the treatment of IBD, which is a little bit sobering. There are technical reasons why it may not have worked. Maybe that the local drug levels weren't high enough. There's all kinds of reasons. But I think the consensus feeling is that you're outside of TNF, probably gonna have to target multiple cytokines. So is there precedent for using the kind of data that you've presented to tailor multi-drug regimens for patients yet? Again, it's, I think, there's no question that the major IBD trials have established that multi-modality therapy, so 6MP plus anti-TNF is more effective than either drug alone. So clearly, I think that we are looking at a scenario where you're gonna have to target multiple pathways, and that's just in terms of the combinatoric complexity that that ensues, it's gonna be enormous. And so it's not clear that it's gonna be a simple targeting pathway like in cancer. And do you think it's gonna be a set of targets for all patients with IBD, or a set of personalized targets for each patient with IBD? So a raging debate in the IBD field is, so TNFs, I don't know, they were introduced quite, anti-TNFs were introduced a long time ago, and we still don't agree. If you ask 10 clinicians, you'll come up with 10 different answers as to are all IBD patients responsive to anti-TNF? And kind of the traditional measures that we use in the phase three trials are very gross measures of well-being and stool frequency, and it's not getting that specific. And so we still, you know, I think that's a terribly important question is every IBD patient a quote, TNF responder, and we don't even have the answer to that. So I'm not terribly optimistic that we're gonna be able to use genetics to subset folks. You're gonna have to have a huge intermediate phenotyping component. So just a quick comment based on Tricia's suggestion in terms of could we have a cohort or a study that is the next place to go after you find your variant? And that's very attractive in terms of the 1000 genomes model where people next went to the sequence data, but the sequence data could be represented basically uniformly across the sequence. I suspect that disease and phenotype and exposure and medication data and electro- and medical record data and that can't be represented quite so easily and readily for various cohorts. So maybe what we would need is someone for them to consult, to interact with and, you know, look at your, you know, a million men or a million person cohort and tell me if there's anything related to this rather than having it be more of a passive kind of just go in and, you know, scan. So it's something to think about at least. I don't know if Tricia, if you have any thoughts on that. All right, we should think about that while we eat lunch. So we've, yeah, while we get lunch and then we'll eat it and we've come to the end of a very productive and I think a very insightful session this morning. We want to thank the speakers once again and all the discussants for bringing, you know, a number of very important issues to the table and some really quite remarkable and exciting stories. It's just really quite, quite astounding. So now what we're supposed to do is go into the next room and pick up our lunch and come back so that by 11.30, Daniel will be at the podium ready to speak. Well, sorry to proverbially he'll be at the podium but he will have munched his way so he should be able to go first to eat if that's what he wants. So, Daniel, you get to cut the line if you want. I should let this up for a minute to tell people we're doing it ourselves.