 Okay, so now I have lots of people here. Hang on one second. So I had Wendy and I have Leah. So but I want to now move into the general discussion for this session on thank Melissa and Nancy for really thought provoking talks. But I want to back up to the people that were in queue. So Jose and then Peter and then Calum and then Wendy and then Leah. My question was related to one of the first slides of Melissa in which you show that change from 20% of explanation to 80% adding the animal models. Do you think that this is just redundancy within the human genome to cope with those changes or with those mutations? I do think so. I mean, when I include the GWAS data in that value it goes up to about 40% but then those are less causal mutations or rather association. So I think there's a variety of things. There's tolerance. There's which mutations do we actually have? So in the models there's often, there's a lot more full gene knockouts as opposed to single variation types of studies in the more historical data that's changing over time. But so I think, so there's a number of different reasons why that's the case with some of the studies that we heard about this morning. Obviously, we're gonna start closing that gap much sooner. So yeah, kind of spans the spectrum. Yeah, awesome talk Nancy. I just wanted to add that we did a pilot project in where we use text mining to get HPO features for common disease. And then we compared GWAS hits which are mainly upstream or downstream of genes with Mendelian diseases. And there's a massive overlap if you just say, can you find one HPO feature that's common to both? So that would also support what you're showing. I think within every sort of common disease consortia, there's an emerging observation that the Mendelian subtypes that there's often common variation affecting the regulation of those that you see that you can see associated at the GWAS level as you build out to larger and larger sample sizes. And we're definitely looking at those same kinds of things. But of course, when there are many phenotypes, they can be associated with multiple. Just really quickly, because it's related to both of those points. So the other thing that we didn't really talk about was the use of the Interactome data or pathway data. So to cross walk across those phenotypes, so if you see connections between those and we've tried, we've had some efforts to integrate those as well. Sorry. From Cal. So I was just gonna tie a couple of comments together. So for example, the Palmer hyperkeratosis is associated in syndromic disorders with a risk of sudden death. If you actually start to measure Palmer skin conductance in everybody with heart failure, it turns out you can actually identify a quantitative association with ventricular arrhythmias and baseline populations. So that sort of highlights how having quantitative rather than semi-subjective is actually quite important. But also that the semi-subjective can inform where we need to go for some of the proximate phenotypes. And I actually like Nancy's concept of axes. I think you can imagine that there are some very clear thematic axes that we've, if we were able to develop translatable phenotypes in that small space to start with, then you can imagine doing everything from the massively parallel mutagenesis all the way through to massively parallel clinical phenotyping with the same end phenotypes. And that's, I think something that might tie the whole thing together, or at least part of it together. Great. Wendy? So I'm following up on some comments and questions for Nancy's presentation. And I think before Howard mentioned it, I had the word Bonferroni in my mind. But I think he answered that great. I'm not one to question. But the zinc idea, the hypothesis, certainly ties in with a way to test this biological plausibility. Because obviously I think you have access to those patients. And to the extent that maybe some of those phenotypes are modifiable in adult age, you can sort of talk what, I think Callum was saying earlier, is you have a biomarker and then you have a treatment. And then you have etiology. So the skin conditions are the obvious place to look first because I think those probably will clear with zinc supplementation, if that is indeed the driving biology, probably the diarrhea, gastritis too. Now whether brains and adults will be as amenable to the therapies as brains and children were, it's hard to know. But I think clearing the skin is a really good first start now. Yeah. Nancy, do you see, because the G-Tex is done by RNA-Seq, are you seeing any missense mutations in those genes that are also associated with these phenotypes that could be less serious or maybe a little, like just the protein is just a little bit broken. Well, so one of the things that we're hypothesizing, so for some of the Mendelian genes, we see really strong additional phenotypes. And frankly, in some of the Mendelian disease genes where there are particularly common mutations and we wonder whether this is kind of an interaction between being a carrier and having reduced expression. So we'll try to get at that through sequencing. Because G-Tex is G-Tex and that's where we build the predictors, I mean, we don't have transcript levels in the patients and subjects, but yeah, you can sometimes pick up the, and we've got whole genome sequencing. I mean, we'll know who's carrying mutations in any of these genes also. So Nancy, I had a question. So your presentation was kind of like a Julia Child show, right? You gave us this beautiful end product and demonstrated how to use it, but I'm wondering about how you put the ingredients together so that the fundamentals of the data integration, how easy was it for you to pull together the different kinds of information you needed to build the resource? What were the challenges that you faced? We'll be building it for a while. So I mean, BioView exists and the, you know, Vanderbilt has a 41 member Department of Biomedical Informatics. So the phenome data has been through a lot of layers, even just in the synthetic derivative. And the opportunity to, you know, we can go across these FIWAS codes and for any of those we can dive deeper with 95% sensitivity and specificity diagnosis. If we wanna be sure that this is dilated cardiomyopathy or primary pulmonary hypertension showing the highly significant associations and what was the particular acute kidney failure associated with the zinc transporter, those kinds of things, you know, we can work with 41 member Department of Biomedical Informatics to get highly specific diagnostic algorithm codes together. But so there was, so a lot of the infrastructure was there. The rest of it, you know, really comes through GTechs. We've been working in GTechs for a long time to develop methods to integrate data from GTechs with genome variation to give people better tools for making discovery. And this is what, you know, sort of just the latest example. But I think applying it in BioView where you've got the whole phenome is a really special and attractive application. I think the sad thing about that is not everybody has a 41 member Biomedical Informatics department to work with. So I know, right. So, you know, so part of it is getting to, what are the lessons learned that could stimulate activity initiatives across the community and make doing what you're doing easier? Kind of like, so, you know, Melissa was talking about standards for encoding and exchanging data, the need to develop that. Do you have any insights into those kinds of recommendations for how to make what you're doing easier for the broader community? I think one of the things that we really need to work on and, you know, maybe with, for example, Melissa's group is the ways that physicians can describe the phenotypes they know about with patients so that it's easier to see, you know, what possible genes come up in the context of, you know, first figuring out what the genes might be. And, but even those, you know, getting the phenome in is a challenge and that's been something that's been interesting to look at from the perspective of the Undiagnosed Diseases Network and all the effort that goes into even figuring out what patients should come into those programs who can't be just diagnosed based on, you know, what they have. And enriching the ability of clinicians to see potential other phenotypes that might associate with known Mendelian diseases, I think, is key. But getting that information readily available in forms the physicians can use, this is gonna be a challenge. But I think there's some things we can try to do that in that direction. Melissa, and then Mike, and then Cecilia, and then, so just a couple words on that. So one thing is that, you know, in terms of crossing that translational divide between the basic science researcher and the clinician, you know, so that we and others have tools to try to help do that, they're imperfect. We definitely need more user interaction and development on those things. But I think, you know, more than that is tools to understand the fact that we are simply never gonna record phenotype data in the clinic at the same degree or in the same ways as we do in the model organisms. And that's where the ontologies can really help with visualization tools. It can help see that this patient is similar to this worm because, you know, clinically, if you're reading PubMed, it's really hard to look for those similar phenotypes. But if we have better visualization tools to help, you know, immediately show, yeah, that actually is kind of similar. Then that's another way that we can do things. And then the second point is that, you know, many of the people that work in our kind of space, the 41 bioinformaticists in your department, you know, we're often at a last, a lack of access to patient data, to sample data, to try the algorithms and try the tools and clinicians who are willing to try those things. Further, you know, we go about our business and, you know, building these data pipelines and data structures and data models. But then the next person goes and builds their own ones that in a slightly different way and has to rebuild that road every time. So how do we better share the process of making these connections, not just the fact that we have the data? Great, Mike? One of the things that I heard from Nancy that intrigued me is this idea that protein coding and delians may also be interacting with regulatory variants for the same gene. We've been talking about second site mutations or modifier mutations or variable penetrants at this meeting and it comes and goes. And that just reminds me of the initial characterization of retroviral oncogenes where initially people looked for the protein coding variants, then found some of them were regulatory only and then, surprise, surprise, found out for some both regulation and the protein coding mattered. Do you have a sense now of how common those three cases are? So we have to really look across all tissues. I think some of the genes that have low coefficient of variation where we don't tolerate a lot of variability in its expression. Also, even if it's highly heritable, we may not get outside the range very much. And so for a subset of genes, we actually see no phenotypes associated with reduced expression of the gene. It just may not be tolerated. But for those genes, there are some bad things if you have too much. And I think that was one of the, to me, a little bit surprising observations. When we pull together the genes that look like Mendelian genes in weighting, there's a couple of dozen of them that are already Mendelian genes with really rare recessive phenotypes associated so with loss of function, but increased expression of those same genes is associated with congenital anomalies, intellectual disability, and other bad things. And so gain of function mutations, or for example, structural variants that might take out regulatory elements for those genes that would lead to their up regulation will look like a Mendelian disease for sure. Cecilia, name Barbara. I wanted to ask Nancy, going back to the comment you had about specificity and sensitivity. And also, I guess Howard had a question or comment earlier with regard to how specific is the diagnosis or how accurate. So I'm wondering whether, I think the tool you presented or the integration of all this data is really powerful. Is there any way to show what the specificity sensitivity is moving forward so that one has an idea, what's the overall accuracy with the approach? Well, so, okay. So I presented two genes that have very dramatic associations. I could have presented thousands of genes that had no associations. So I don't want you to get the idea that I got a lot of genes at 10 to the minus 19. I don't have a lot of genes at 10 to the minus 19. And yet, unquestionably, the Mendelian genes do attract a lot more phenome. Much of the phenome they attract is the part of the phenome we already know should be attracted. But there's some new phenome also. Some of it does come in in these axes. So I think the axis that the zinc transporter is on, this wound healing and native immunity, it involves kidney failure, it involves some of the primary pulmonary hypertension. These are things I've seen with a number of genes that show associations not as dramatic as what we just saw, but significant, you know, 10 to the minus eight, 10 to the minus nine. And so this notion that, you know, we sit on these planes and rock back and forth, depending on the environmental exposures and the whole set of genes that we have, I think is real and that there's a lot we can do there. With respect to, yeah, we're looking at high throughput zebrafish, cell-based assays, and mouse studies with a lot of the novel findings. But I'm not gonna take, I mean, there already are mouse models or sometimes zebrafish models and drosophila models even for some of these Mendelian disease genes. When I see the same phenotypes associated with the Mendelian genes for reduced expression as with the disease, I think the model's complete. And I'm looking at the novel associations for validations and we'll have some quantitative description of how many of the novel genes with associations that we put through this we can validate. But again, you know, with strict Bonferoni correction, a lot of these will definitely meet that criteria and we don't have that many that are so dramatic. So we'll go to Barbara and then Bruce and then we actually have to wrap it up because we have Bob online for the next talk. I'll be quick. So Carol, you mentioned, you know, how applicable is this to people who maybe don't have access to a bio view and there's something that Nancy didn't talk about really with relation to the imputed gene expression and that's that if you have whole genome sequencing or if you have genotypes and cohorts, you can impute the expression in many tissues that have been surveyed by GTEX and so you could imagine in a case control study you can impute cases and controls, find out which genes are genetically predicted to be differentially expressed and learn potentially about causal mechanisms for individuals you don't have the transcriptome data for and that wouldn't be affected by their other phenotypes and those models exist, those models exist, you can go online and that's all publicly available. All publicly available. Bruce? Yeah, so there's been a lot of discussion actually just in the last few minutes and for that matter even in the last day or so about all the things we wish clinicians would provide in terms of phenotypic data but they don't and it just strikes me maybe speaking in a way on behalf of the medical genetics community which could be the source of a lot of this, there's a small handful, probably many of whom are in this room who think in these terms and then a very large number where that's completely foreign territory to them but I think you would find actually that there's a high level of interest and willingness to participate and I guess I would just propose somehow that we come up with a way to reach out to that larger community and try to engage them and I think you would find a lot of willingness, there's not that people don't want to put phenotypic information in and I would even argue actually some of the discussions about IRB or other constraints are somewhat overstated actually I think. A lot of the patients I deal with would love for us to share their information in the hopes of coming up with better data but my point is that I think there is a large small army actually is probably not large but at least there's a small army of people who would be willing to participate if we could reach out in a way that would break what I think is a bit of a tower of babble of all the various databases that we've been talking about. Well and maybe I could just comment in response to that Bruce, hoping that you and others, Gail and Mark and Mike and those who are heavily tied into the ACMG community will stay if you can through the discussion in terms of next steps because one of the things we see as really useful would be going back to the ACMG and really trying to make these links in ways that make sense to the average clinical geneticists not those of you at one end of the spectrum that are here with us. I just wanted to make a sort of an observation of a paradox and the people sitting around the room who come from the Model Organisms Community will know that there's been a lot of concern lately about how Model Organisms are gonna be funded going forward, Eric and other NHGRI folks this is for you as well. So I think today we've seen a really great demonstration of the value of the data that's collected by those databases. So I would just urge that we make sure that we don't break something that's working really well going forward. Yeah I wanted to add on to what Bruce has said. I mean one of the issues with any requests from clinical data that we've heard several times already is what's the value for the additional work that I'm going to do. Now the reality is that in the context of ordering genetic tests, many of the companies are now requesting relatively sophisticated phenotypic information to go with the test. And we also have heard some presentations about how through the standardized ontologies in that there are actually ways using check boxes that you can kind of fill in what's present and what's not present. But we don't have across all those things as an integration and so you could imagine, I could imagine, a scenario where if I were in the clinic and I had a patient and I said, gosh if I fill out this form, it's gonna go to the laboratory with any testing that I order. It's going to allow me to immediately pull up relevant information from all of these resources and it would facilitate bi-directional communication and could in fact under appropriate permissions be deposited in something like a monarch or a ClinGen as patient level data. If you could do all of that with filling out one form, well now you've really got something because I get something from doing it because it makes my life easier. I'm not having to go and do individual searches of all the resources and the value to the community is evident. So that again is another sophisticated informatics solution and interoperability problem. But if we don't get the value proposition right, other than the few altruists which don't represent even a small army, you know, we're not gonna make much progress. All right, Wendy was giving. I'll be really quick because I'm just building on what Bruce said and what Mark said and I realized we're drinking the same Kool-Aid for a long time but just to build, this really is about a value proposition and I think all the tools are there and looking in part of the value proposition is getting accreditation, of course, and looking at Mike. But I think that it can be even better than that because it kind of comes down to who pays for these genetic tests really? And can you even order it? Which is kind of a big no in many circumstances. So it means that we have to partner with the payers and I think we have to go beyond CMS to do that but there is increasing interest in the payers participating and FDA participating in this whole endeavor which I think we can leverage. Now it's not too much, I think, to ask a clinician who is ordering a genetic test to enter in one, two, three, four phenotypes that become part of this collection and then the benefit for the payers is, number one, you have a narrower gene list as long as you're not in the discovery phase, you're talking about genes with known phenotypes and then you can get that to be iterative once you have a result back and you have a variant that pathogenic, VUS, but you can recategorize it. So I'm just saying to extend this discussion to really the whole kind of ecosystem of people that need to help us to build it. So this is really awesome, I hate to do this. We do have a discussion, we have a panel, so those of you who didn't get to comment right now hold the thoughts, bring them back up when we have the panel discussion, we have plenty of time then to continue this because I think this is really important trend of discussion. But now I really wanna turn this over to Bob Nussbaum who's online and is not a hologram, unfortunately. And he's going to give a talk which the title has changed to something very interesting just before lunch. And I might just mention if Melissa got the Hero Award, Bob gets the second Hero Award for doing this remotely for us and while it may be even more tempting to scan through your email and that because he's not here looking at us, he's gonna give us a hint as to where he's buried a million dollars and it'll be in his slides, so you wanna look carefully at this. Bob, go ahead. It is floorboards, they gotta flood it. Yeah, exactly.