 All right, why don't we go ahead and start this morning? Thank you all that made it back. As you may have noticed, the tables are a little bit narrower and a little bit more towards the front. So no, your life has not changed that much. It just seems like it. So in case you're wondering where your seat went, hopefully there's still a comfortable and available to you, even if it's not exactly the same as yesterday. We want to go over a little bit of some of the discussion last night in terms of some of the future meetings and such that could spring from this. Also wanted to have some time for anyone that had any big ideas or thoughts that need to be shared in terms of yesterday's program. So why don't we start with that? If there's anyone that has anything major to add that just was not brought up from the six sessions yesterday, I'd love to hear it now. So I don't know, take it. I just want to, I guess this is more in reflection on the priority items that were discussed yesterday. I'd like to just throw out, under the evidence one, the concept around deep phenotyping. I know we brought it up a little bit, but it is what I would call a high hanging fruit and it's one of the harder ones to actually deal with. But it is right now one of the major blocks in translating a lot of what we're doing. We're great at finding a lot of these gene signals and connecting them to diseases, but what we're not doing is taking that and actually validating them. So I would suggest that maybe one of the things that we could think about as a priority area would be trying to focus on deep phenotyping. Is that the way to go about validating them or are there other approaches? Or is that the only way? Obviously not, and I think to the degree that we've talked a lot about the virtual cycle and is there a way to bring some of those back to the bench and have them investigated. And I don't know that we've done very well with that and we've talked about it in this group a lot, but we still haven't been really effective at it. And so thoughts about, and perhaps when we go to sum up this afternoon, thoughts about ways to facilitate that kind of work. I don't know, Jonathan, you're looking very eager. So I was just gonna comment that I was on a study section for sort of functional validation of regulatory types of SNPs. And there are some people doing very innovative work in functional validation of Mendelian-like, sort of protein altering hits. And I think facilitating that type of research would be very helpful. I've been engaging people at UNC to try to do something around that. And what I found with the basic scientists has been largely that they're very interested in gene discovery type applications. So they definitely wanna know all of these interesting, interesting hits in novel genes so they could follow those up, but much less so interested in setting up a nice allelic series like the one that Sharon described yesterday from Fergus Couch where we really need to understand all of the variation in this gene that we already know that is involved in disease. And I found that to be a less tractable or less attractive thing for the basic science folks. Adam, I wonder if you could just, we've heard the term deep phenotyping and I'm not sure we all really know what that means. What does that mean to you? So the way I'm gonna comment this a little bit more from thinking about a clinical application or a commercial application of genes and when I'm thinking about it in that sense, I'm really thinking in terms of understanding mechanism of disease action. So going from gene signal to actually understanding the mechanism of disease. So if we can get a better understanding of the underlying biology, the full pathway and understanding, that gives us an opportunity to actually validate targets for new drugs, for instance, or new molecular diagnostics. I mean, that's kind of the area that I'm thinking we need to do a little bit more exploration in. And similar with John, I think we're saying it's not, this isn't the area that many basic scientists want to get into. It's much more in that transitional pathway and understanding how do we actually move down that pathway of development. So at least that's the way I define it. Jeff may actually be able to add, we've been having these discussions in our discovery group, so. Well, I think the term can mean a lot of things to a lot of people, obviously. And so the focus on the mechanistic aspects makes a lot of sense and as Adam said, we've, and we mentioned yesterday, the, you know, we've got members on the roundtable in the pharmaceutical industry that are extremely interested in mechanistic applications of this work. But I think the other dimension that I think some people in this room would be also interested in is when you make a association between a genetic variant and a characteristic in the electronic health record, for example, why not broaden that to include all of the characteristics associated with that patient and as we've also alluded to, begin to bring into the longitudinal dimension of what happens to individuals with that series of characteristics and those genetic variants so you can begin to really create a bigger picture of the implications of genome variation to prediction and to disease association. So I think it has sort of flavors in both the mechanistic side as well as in the clinical translational side. And I might just know that, you know, I think I've heard the term deep phenotyping used much for a much more careful examination of a patient and it sounds like the way it's being used here is more cellular phenotyping and functional understanding, not, you know, let's dive deep into, you know, every, is it, so it's really both? Because I'm not sure that with GWAS hits, necessarily characterizing the individual more carefully is going to tell you anything more about that particular association since they're common variants, they're tagging something and so I don't know if, you know, how well that will work. Well, it also, I mean, the total body MRI could be deep phenotyping, you know, 100,000 metabolomic profile could be deep phenotyping. So there's a lot, it depends on the goals, I guess, but I do think a more definitive understanding of the functional consequence of a variant would be useful and it'd be useful not only for biology, but in terms of, you know, what we talked about earlier, in terms of trying to shake it and see what's real and what's not and so I think it definitely is not, there's a need for more of that. Jonathan, I was wondering, I know Dan's gonna be next, but when your discussions with your basic science colleagues, were they not interested because there wasn't mechanisms to support the work or were they not interested because it just, that's not what they did, right? I mean, I don't know, I can't speak for them. I think partly it's probably more difficult to get something published in a high profile journal if it's, you know, really functionally looking at a bunch of variants in a gene. And so there's that aspect to it, but. You know, one tension that I sort of sense in this discussion is, you know, are you starting with a variant or a set of variants and then doing deep functional characterization of a deeper understanding of what they do, or are you starting with a human phenotype extracted, say, from the electronic medical record and then trying to understand, you know, what set of variants drive that? And variants can be genetic variants, other omic variants, environmental variants, sociocultural variants, what have you. And so I think that precision phenotyping is one of the major challenges to this field. But I'm not sure what the right way to approach it is, whether it's from, you know, starting from the phenotype or starting from the variant. And I suspect that the right answer is that you should approach it from both ways. I really like this idea of, you know, we've been talking about this forever, that type two diabetes is not one disease, but 25 different diseases. And what does that mean? Well, I think it means that there are subsets that are definable, reproducible, and of some clinical utility. Why would you bother to define a subset if they happen to be, oh, that's interesting, that's associated with this variant rather than that variant? If you define a subset, you define it because they're different in some important way to the patient. Their tempo of disease is different, their risk of adverse events is different, their risk of complications, their risk of responding to a drug is different. So I think that there is great value in thinking about that. We've all thought about it that way, but to sort of call it, and put it into the basket of precision phenomics seems to me to be an interesting challenge. And I would start it with it from the functional, from the individual variant point of view and from the individual patient point of view, but getting these, for lack of a better term, clouds of, or sub-clouds of patients that have common subsets of disease is something we should drive for. But that also fits really well to something else we've talked a lot about, and that is this brings us back to the biology. I mean, it's fundamentally about the biology. Yeah, I don't think yet. And that's, and it's important when you talk about bringing it back to the biology, not to bring it back to sort of traditional 1970s kind of biology, but that plus newer kinds of analytical approaches that'll find those subsets. Well, Dan said just about everything I wanted to say, but to reiterate, and the- That is truly frightening. Yeah, I know. We talked yesterday about the granularity of the phenotype, and as Dan was saying, 25 flavors of diabetes. Well, to the extent that those flavors are therapeutically relevant, or prognostically relevant, which is what you were saying, then I think there actually is utility. Teri Monoglio yesterday dinged me somewhat for, correctly, for saying that, well, there's actually a lot of resilience in the phenotype definition, particularly with GWAS association. But I think as we discover more rare variants, and as we discover more heterogeneity within disease, that we'll find less and less of that resilience, and that the importance of, whether we call it precision phenotyping, deep phenotyping, whether it's at a, you know, a molecular physiologic level, or whether it's at a clinical manifestation level, it's still a granularity of phenotyping that I think is going to be increasingly important. And I might add, just for good measure, Mark, that to the extent that once we have capabilities and reasons to do these deeper phenotyping, then the consistency of how we represent clinical data is going to matter even more than it does now. One of the challenges is we've got a scale effect with regard to the sequencing, and we've identified more biologically interesting things than we've ever had before, but we don't have a scale for following those up at a molecular level. The biologists, you know, they can spend 10 years following up one thing, and yet we've got a million things now that they need to follow up. And so there's a gap in the technology at the molecular side at the moment. I think it's improving with some of the new techniques and technologies, but that's the side that's going to bring this along faster and help us get to the biology faster and link these things into the pathways they're in. Yeah, I think that's an interesting point, and then it raises a different role, perhaps, for NHGRI in the sense that if we're thinking about this sort of bidirectionality or virtuous cycle, however we want to characterize it, if we, in fact, have this target-rich environment of a million things to do, then how do we focus the ones that are most important and prioritize those and get people interested so that they can do the functional studies or whatever? And so what that really reflects is some type of a rapid filtering and prioritization to say these are the ones that we really want to focus on. I might ask my colleague, Mike Payson, to comment a bit on the issue of scaling these things because doing them one at a time is, you know, not feasible as you suggest, and there has been some very innovative work in ENCODE in the past couple of years looking at all GWAS hits at once and saying, you know, gee, there's an awful lot of these that are at DNA-sensitive sites and that sort of thing. Mike, do you want to comment at all? Yeah, so first I'd agree with what I think I'm hearing that today. It's easier to find these candidate associations between variants and disease than it is to understand them. And for some time it's been outside of NHGRIs considered it outside of our purview to follow up on the functional characterization of variants that are found in disease studies. Something that was said openly and available to anybody that has internet access at recent council meeting is where we'd like to renew our functional genomics program and we'd like to add in a new component that would include high throughput functional characterization studies. So those RFA's are not written. I don't know for sure what they will look like and I haven't seen there for any of the applications but it's conceivable that some of the work could move in this direction as sponsored by NHGRI. And Mike, do you see a role for interactions with clinicians? I mean obviously there's a role for interaction with clinicians but what role could you see and are being able to either influence or help to prioritize that sort of thing? Yeah, so in the preliminary discussions that we've had on this, that role is not well defined and I'd say on my part it's not well thought out but we definitely would like this project to have experts in biology, experts in disease as part of the project. So I see opportunity, not well thought out, not well defined again on my part but there's opportunity to work with partners in other areas of science. Maybe Craig just to challenge you a little bit. So say you had some colleague that wanted to be involved in large scale characterization of function and you've spent decades studying diabetes. How would you help them to sort of come up with the biggest bang for the buck in terms of that disease? Some goes back to the deep phenotyping and this interrogation of genes we've done and the identification of, we'll call them tolerated mutations is such a rich resource that generating additional copies of those through pedigrees or other kinds of things I think provides them a way or us a way of prioritizing some of those. Why aren't these having bigger effects than they have? It's partly because the level we look at but I think more interrogation from a pedigree point of view that's a very efficient way of getting at additional mutations and then there's so much that we know if we just look at the insulin gene. 41 of the amino acids, there are natural mutations that change amino acids that are semi-tolerated. So you can live with them with some treatments and I think identifying those and moving those straight into a molecular pipeline or a functional pipeline will do great things but we can help move those both by what's tolerated and by what we can see at a lower level. Right now we have things that must be having an effect but we're looking at their effect on diabetes as opposed on the flux through a pathway and its influence. And I think that's back to the issue of what deep phenotyping is. I just wanted to add, I've been thinking about this a little bit with ClinGen and I know we've had some preliminary discussions but as we're spending a lot of time assessing variants and looking at the clinical validity of genes, would it be worthwhile for our group to figure out ways of prioritizing and we're really pulling together all the evidence now and identifying the gaps and the variants that look really good and the ones where we need a lot more information on. Do you see a role for ClinGen in this space? Yeah, I mean it's something that's come up a lot as we've tried to develop the sort of standardized ways of evaluating functional evidence and I've Sharon mentioned this yesterday, right? You come across a piece of functional evidence that's been done in a well done fashion but it doesn't quite reach the level of a clinical test to judge whether that variant is truly causing problems and so I think there would be a role for ClinGen to point out places where new functional evidence could be useful, right? Here's a whole group of variants in a gene that could potentially be called benign or called pathogenic if we just had this additional piece of data. Well and I wonder too if it might be very useful, it would be probably pretty easy to say of the 20,000 genes in the genome, 19,000 of them we know so little about but really to kind of find those that almost as you said are kind of teed up what's been described as the half-baked bread and you just need to put it in the oven a little bit longer and if you could figure out where those are or identify G if we only knew X about this particular variant in epilepsy or whatever it might be. That I think would be very useful then of course you need to find the basic science colleagues who will look for X which may not be an easy find. We also have maybe a resource among all these consortia and groups that have been listed here and are assembled here because we've sequenced a lot of those, we phenotype a lot of those and they're already signed up for research projects so I wonder if there's a way to leverage some of that to go back into those individuals and re-enroll them in physiological studies and re-phenotyping, deep phenotyping, precision phenotyping kinds of initiatives. I also wonder if there's any value to thinking about it the other way around. As we get, we clearly don't have enough today but as we get more and more sequences very broadly across the people I wonder if there's also an equal interest in the variants that are never seen. Sort of the holes in the genes that for some reason are never touchable. Those seem to me they may actually be particularly useful for a more basic biologist that go back in and say, well here's an easy way to think about doing a knockout especially with tools like CRISPR around, so maybe building a resource that says where are variants that we've never seen. Yeah, I mean that certainly has happened many times where someone has identified a disease gene association in the mouse and then a month later it's published in humans. So I think there is that discovery starting from the model organism going into disease but we have increasingly the kinds of virtuous cycle examples you were talking about yesterday including in epilepsy and diabetes and other things where the variants from sequencing in human populations are identifying priority variants and then through CRISPR go back and make the mutations in the mice and the nice thing is you can do more than one gene, you can target whole pathways, you can really perturb the system in very directed ways to make that functional connection between variants and biology variants in disease. So I think we're primed for this sort of virtuous cycle. It's just how do you get the right people talking to one another that will develop new tools, which NIGRI's mission is not to do the fundamental biology but maybe to develop new approaches and new tools that will scale that up and I think we're in a prime place to make moves in that direction. This is more of a question related to something that Robert said, I'm wondering are the sequencing programs collecting other types of samples that could be used to explore cytokines, metabolomics, the transcriptome as endophenotypes or to begin to tie the genetic variants to some of these other biological networks, would that be of value? It's a great point. Cesar's not but we certainly could. Yeah, we're doing that, Jeff. I really do believe that that's a great way to just leverage resources. It costs so much to identify these patients and roll them, genotype or sequester genomes not to do that seems like a waste. So we've been doing that, albeit we haven't, we've just been collecting samples and getting consent. So I think that the whole biorepository stuff raises yet another metadata informatics question. So you might have them in your database behind your firewall, but how do we, if people are developing these biorepositories for these samples, how do we actually then annotate them and put them online so that they're shareable? That's broadly, that's a whole nother data format standard, semantic standards issue that would have to be dealt with, but certainly doable. Yeah, and as opposed to the fact that we've declared victory on the storage of the genomic data yesterday, the reality is a storing biological specimens is a huge issue. And so if we're talking about scaling that up to collections on tens, hundreds, thousands, millions of individuals, there really isn't the capacity to do that at the present time. I mean, the advantage of DNA is that you don't need a lot of space to do it. And in fact, filter paper works great, but for a lot of the other things you can't. And then the question is, is there a way to prioritize which things would have the best yield? So much as Stephen is doing in his project for certain of the projects that we're doing like in our obesity institute, and then we're gathering serum and urine, and we're actually, during the surgery, we're getting liver biopsies and adipose tissue, which I think is really relevant for that obesity phenotype, but probably is not so relevant for other things that we're thinking about doing. So is there even a sort of a standard set of biological specimens that you would want to collect and what would be the highest utility? That could be a question that could be studied. So thinking along those lines, I'd like to ask how many groups are collecting RNA samples in their work or considering it? One of the advantages to coding mutations, it's more straightforward to predict what's the target gene and then follow that up. But of course, once you've found that, it's hard to know whether that change in the protein is important or not. The downside of non-coding variants is often you don't know which gene they're connected to, but if you do, you know it's probably changing the expression of that gene, not the function of that gene. So what's nice about that is if you know that a non-coding variant affects gene A, you can find that difference simply in the RNA then and say, well yeah, the expression went down and the expression went up. And then with a single assay, you could look at many genes at once, it's a nice genomic approach. I don't know that any of the major genomic medicine programs are collecting RNA. I think maybe a couple of the CSER programs, remind me, Lucy isn't Chenayan doing some. I believe Michigan, the Michigan group is. That's the one that comes to mind off him. The problem of course with these others, I mean, there is a reason that we've all started with DNA and that's because there's no temporality, at least as far as we know, there's no temporality with DNA. Whereas with all of these other samples, it's not only, A, and we've really looked into the logistics of what does it take to get a serum sample that's actually of value if you wanna do metabolomics, for example, there's a whole problem of how long do you have to get it into the freezer. And then there's the much bigger fundamental problem is, are you sampling at the right timeframe in the biological space? And that adds so many variables. That's not to say it's not worth doing, but it certainly boggles my mind. And it probably would behoove us to think carefully about how and when we would draw those samples so that we would get the best functional information. And again, maybe that's something that we need to discuss in collaboration with those who would use those samples. Yeah, I would just point out that one of the things that I think has been exciting in the last several years, especially in space that I know best, which is places where we're actually linking samples to clinical records or to well phenotype samples, is that those samples actually get used by, they get used for experiments. And the problem with a lot of biobanks in the past is they were collections of samples, none of which ever got used. And so, and it's a very expensive process to collect samples and never ever use them. Well, many thanks to Adam for starting this conversation. Clearly there's a, no seriously. It's obviously it resonates. And it happened to be one of the topics we were talking at the steering committee or whatever it was, the working group, whatever we are. Last night about what's next? And there's a number of next issues that have been brought up yesterday. And one of them was exactly what we're just talking about now and that is how do we bring, you'll really create this circle that I'm not sure how virtuous it is, but that is needed and bringing discovery science back into a lot of what we're looking at. And so I don't know, Terry, you wanna comment a little bit on? Well, so yeah, so one of the things that was raised was gee, we have an opportunity, I think, to bring both basic and clinical scientists to the table in a forum like this. And I'll ask Carol to comment as well on how we might best do that. But it seems as though if we could perhaps invite some of our leading basic science colleagues who are doing work that they feel has clinical applications or could be informed by clinical information, they would be good people to have at the table. And then I guess in some ways we would describe sort of what we have in terms of resources. Plus what kinds of questions are we stumbling up against that we really need help basically from? So could you comment further? No, I think that's right. I think the starting point is to have some folks come who are already trying to do this, the same kind of virtuous cycle thing, have them come as exemplars, sort of describe what the process is, what the barriers are, what the gaps are, what the promises and what the results are, and just sort of use that as a launching point to figure out how to make this, the uptake of this approach more broadly applicable. I think that's the way to start. Start with some small examples where success is obvious and go from there. And then one of the other discussion points that Jeff had brought up was bringing in some of the industry players, like what we're generally done with Geisinger and show really there's a new found notion that if you think of the human as being the discovery start point, the basic research then is actually the follow through, which is a little bit opposite as Carol was saying. We used to do in the animal models, we find the gene there and then try to see if it plays a role. But industry, academia and all these things are now moving from human back around. So I think industry that's doing this maybe Regeneron is one of the examples. Yeah, I would like to advocate to go back to what we had done for GM 1 when we were trying to really organize where we essentially solicited send in a little description of the projects that you're doing in the implementation space for genomic medicine, so that we really have a pretty broad sense and then choose from there what we think are the most informative programs for actual more formal presentation and discussion. And I think that that would be a really nice way to kind of explore that space. Just one addition, once again, thinking about linkages north and south of the border. But my colleague in CIHR, Paul Lasko, who you probably know well ahead of the Institute for Genetics there, has set up a network of looking at functionality of gene function in general for looking at exactly the things that you're talking about and building a network of basic scientists working in yeast, Drosophila, mice, zebrafish and so on. And that network might be an interesting linkage very much like-minded thinking, I think that's going around this table. And that's up and running. Both CIHR and Genome Canada are putting money into that. And Genome Canada has just funded a new platform on Phonomix at the Big Mouse facility in Sick Kids in Toronto. And they're going to be looking at it as well. So it might be some cross-border stuff. Is there a name for that network up here that you- Yes, there is, but I can't remember what it's got. I've forgotten that gene that controls the- So I don't know why I forgot about this, but the office of the director has a U54 program that they're going to make decisions on at council next week on a precision animal model grant. And I think they're planning on two pilot centers. And the whole premise around these centers is to actually do this work starting from variation in human and then being able to model it and take it back to human. So I think looking at those programs that end up being awarded would be another way to bring this in. So we also talked about GM9. Oh, that was GM9. And then there was the scientific meeting. Yeah, do you want to- Do you want me to- Yeah. So sorry, we're getting our act together this morning. Sorry, the Tarian Howard show. So something else we discussed last night was the richness of the discussion around the table of, you know, we can pat ourselves on the back of the Genomic Medicine programs that we heard about and others that maybe we didn't have time to hear about. And wouldn't it be cool if we could have a scientific meeting that would really be devoted to sharing information about those programs and discussing them in a scientific forum like a Keystone meeting or a Cold Spring Harbor meeting or whatever. We're kind of wondering what people thought about that. As you know, none of us needs more meetings to go to, but we- And part of it was, look, it was definitely not to have another meeting because if, and we even talked about, well, if we're going to add one we need to subtract one somehow. But there's a lot of opportunities across even the programs that are represented here that just don't happen. You know, there wasn't a speed dating session at this meeting, for example, not that we want to have one, but the idea of, can we create something where there is these interactions that happen with or without further funding that goes with it? Yeah, I know, I love the idea. And I think that for somebody, this is my first meeting here and UDN is one of the newest, maybe the newest program. And just listening to the discussion and noting what's happening in other programs that actually I was not aware of, maybe I should have been and could have been, has been really wonderful. And I think that if we were going to have such a meeting, I'm sure you would do this, you would organize it around certain themes. And that might, you know, the deep phenotyping, for example, might be a theme that we could bring in the kinds of different points of view on what that would mean. And so forth. So I love the idea and I think it would be great to have allowing for plenty of discussion in the plenty of discussion rather than just a series of PowerPoints. I'll just throw that in there. Yeah, I want to say from the Caesar perspective that we think that would be a very valuable idea as well. And I also want to mention that we do invite people from some of the other networks to attend our in-person meetings. And we would be glad to go to the Ignite meeting or whatever, other Emerge meetings or whatever to talk about Caesar as well. And so we have had those kinds of interactions. I just want to ask not to do it around ASHG because I already have two weeks of pre and post meetings. Well, let's make it a full month then I guess. I think some of this came out of a discussion around, you know, Ignite having, you know, Caesar coming and Emerge and Ignite and Caesar and Ignite and Emerge and blah blah. And how fruitful those have been. And with the number of groups, it's almost impossible to do that in the course of our lifetime, you know. So can we create something that would expedite that? So that's great. I think Tampa in February was what we were talking about. Other comments? Just to comment, we're going to let the, you know, because we do tend to lose some people, some of, none of nobody here, of course, a little bit later in the day. We wanted to be sure to have a robust discussion this morning. And I think too we can let some of the panels go a little bit longer than our if the discussion is robust because really what we need is the discussion amongst us. It's what we all know, not what each of us knows. And so the, you know, the recap at the end could be short changed and we can do some of this follow up, you know, by email in other ways. So. We were talking about the survey. Right, yes. Why don't you mention the survey? Okay. So again, one of the things about summing up the meeting is to try and get a sense of prioritization and we recognize that as we had evidence of this morning that not everybody can sort of instantaneously synthesize, you know, two days of meetings and, you know, come up with a reasonable prioritization. So we're going to use something. Sorry, we didn't warn you guys, I'm going to use a little bit of a strategy and read a by and Jackie about this, but we're going to use a process that we did for GM 7 where we had a post meeting survey that goes out that basically takes sort of the topics that we have agreed upon as a higher priority topic. Send those out. Have you kind of ranked them in a survey monkey or something of that nature? And then we'll use that to, in the final analysis of the meeting. If there are themes that we didn't pick up on or as on reflection you think are more important, there'll be opportunities to enter those in text as well. So we'll be doing that. Yeah, so I have to, keep looking at me, you know, it's like, and now I'd like to introduce Howard McLeod, who will be presenting panel seven, Streamlining Clinical Workflow, Transportability and ClinVar Submissions. And so presenting from there, go ahead.