 Yes, so I'd like to invite the speakers back to the table. And I just asked for her to bring back the slide. I think it really sets the stage very nicely for discussion. The many of the challenges and opportunities that were had been identified in setting up this panel. And do you want to say anything? No, nothing to add. I know it's after lunch. I hope that everyone came with energy for one more really good discussion. The discussions this morning were just phenomenal. So hang in there for one more time. And yeah. I was going to make this actually perhaps get started discussion referring to the point about high-quality reference genomes, because it was discussed in the morning. And one of the things that comes to mind to me is while we are looking at, well, we had a great session. There was no major talk about mouse genetics. Pretty amazing. Look at the potential of non-traditional model organisms or model systems. And then the need for high-quality reference genomes, I was thinking back, it's great to have that near-perfect type of genomes. But there's probably limitations and constraints about which ones would be selected. So try to think about opportunities for non-traditional organisms and perhaps having to find, or in terms of guidelines for funny agencies, about some more pressing organisms, model organisms, or something I keep thinking about. How do you go about, and at least among the points of the discussion, what are the interesting phenotypes? Are there the pressing things that could be targets? And I was thinking about also high-reference genomes for those organisms. So just like that's a point of the start of discussion. I mean, I'll say something, but yeah, there's a comment. I just want to get a clarification. Claudia, are you asking like prioritization for investing in funding of organisms to do genetic manipulations? Yeah, to add to that point, for example, the NSF is funding the edge approach. We are part of one of the edge things to develop tools for manipulating, for doing genetics and the manipulations in non-traditional organisms. And oftentimes in those organisms, there's not high-quality genomes available. So it's kind of thinking about the interplay of how do you pick high-reference genomes and to leverage the opportunities for looking at interesting phenotypes that are not necessarily present in a mouse, for example. And I'll think about those things in terms of to maximize the genotype, the phenotype questions, the opportunities for those. That's kind of coming to mind. So Rez? I mean, I think a reasonable way to think about that is, are there particular clades that could accelerate everybody's work across USDA, NSF, and NIH? I mean, I think certainly for us, we've thought that a sweet spot is the mammals. I know in Eleanor and Shustin's work, that's been a focus for them as well. So that may be a starting clade, but there may be some other clades that are really critical. For instance, an ag, there probably are some pretty obvious clades that we should look at. I'm going to guess my comment to that would be that we all, well, I have lots of pet organisms that I want the genome for. And if there were infrastructure resources, like the EDGE, but maybe on a smaller scale, to say here there are these calls for $10,000 for your genome, here are the ways that you have to do it to fit in the vertebrate genome project or whatever it is, and then you can make your pitch for what's the impact, who's the community. And in some sense, it's reviewed. Although I'm not sure that I love that suggestion at all, but it's an idea. It's a great idea. I think fostering or funding for development of general use gene manipulation tools that would work across species, I think would be a valuable way to answer that question. Because there are a lot of people, like Claudia, yourself and myself, we're focused on trying to get something to work in the poor zebrafinch, but it might not work in a woodpecker or a C. elegans or something like this. So if there are people funded to develop manipulation tools that they can test across species, I think that would be a valuable contribution than funding a whole bunch of different species. Well, what about the genomes themselves as well, right? The high reference genomes? Well, since we're going to try to do them all, that doesn't matter. Well, there's a prioritization. You won't get them all, right? I have a very specific question for Dr. Spencer, you showed a slide where you had a series of genetic features that would influence fertility. And one of them was homozygous recessive lethals. Did you actually quantify the number of homozygous recessive lethals or the number of lethal alleles in your cattle or in an individual? Right, so basically that's not my work. It's other people's work, like John Cole and people like that. So did they quantify them and count them up? So it turns out that pretty much every breed has one or more of these embryonic lethal alleles segregating in them. And basically, we've done the same thing. Jerry Taylor at the University of Missouri has done the same thing, where you simply find sires that don't have the right mandaline ratio in terms of their pregnancy outcomes. And so they backtrack them. And I was kind of shocked. I mean, some of the Holstein or the Jersey, the JH1 habitat, for talking the carrier frequency is 10% to 20% of all animals. Did they identify which genes we're carrying them? OK, but because this kind of work was done in Drosophila 50 years ago or 40 years ago when they're trying to compute the genetic load and understand quantitatively how many genes were neutral or selected in those days. But it was the same kind of question. So if there's data out there on specific alleles and the number of alleles quantitatively in these different breeds, that's exciting. Yeah, and I think probably two people have done probably most of the work with the USDA are a spelt spill, as well as Michelle George's in Belgium. So it's really a fascinating rate. So I think there was a question here first, and then we'll go to you. Yeah, so I was just wondering if there was some value in turning the question of genotype to phenotype on its head a bit, because it feels like we have all of these kind of scattered small projects looking at one species in a trait that they've adapted to. Whether there might be some value, and you've got the 13-line ground squirrel doing interesting things with glucose, you've probably got a whole vast spectrum of other species that do weird things with glucose and managing glucose and can actually look at it from that direction. Identify a number of species that are doing interesting things in that phenotype and then start looking at how they're doing it in terms of their genotype. Yep, great point. Back. Claire Gill, Texas A&M. I just wanted to comment on Tom's point about genetic load. And in the beef breeds, the load is about 30 per animal of these deleterious mutations. And it's a problem when you're looking then at pure breeds, and we have breed associations making selection decisions about animals, and they want to eliminate these from the population. So how do you actually do that when every animal is carrying a load of about 30? Was that 30 per individual or was that 30 altogether? 30 per individual. This is Donald Manahan. I had a question or comment, just wondering from the panel before you to this panel. So the panel before you, it's a linking of genotype phenotype. So the panel before you was spending a lot of intense conversation about how to normalize data that others could use, right? So then we go after the coffee break and even just take Joanna's enzyme assays, right? You know, when you look our animals assays or any of the assays of the cell types that we heard, it seems to me that the gap here of trying to normalize all those things are infinite possibilities just beyond comprehension, right? And is it surprising then that we see reports coming into science suggesting that experiments are not reproducible in science and all the public criticisms, et cetera, that come with that? But in fact, they might be reproducible. It's just that it's impossible to reproduce this level of phenotype. So I just want to open up a comment and a conversation about at one level in the genotype world, in the genomic world, we have an intense conversation about, forgive me for saying so, something that's somewhat relatively simple. There's like four letters, right? You know, and then we go into the phenotype, as Joanna said, showed very nicely in her slides, you have biochemistry and physiology and all these things changing with the environments. We made some really new thinking on how to do this. And I think it's particularly urgent, given we know the high level of criticism that science is under generally about the society is confused, it's told to do this, then the scientists say to do that. We understand the nuances of reproducibility, but I can see even in science here, just from these two panels, that there is a massive gap between genotype and phenotype. And I just wanted to make that comment. It's not something we all thought about it, but what kind of new thinking do we need to do to try and close this gap? So I think that there are two different issues, how to normalize data set and how to generalize your result. And I'll start with the latter because that has to do with reproducibility. I don't think there's reproducibility crisis. There's over interpretation of the data sometimes that leads to conclusions that are difficult to the conclusion themselves support. And the easiest example is when you study sample with a very small sample size and you generalize over the entire species. And then somebody else comes with another small sample size and either due to lack of power, because incomplete power small sample size is considerable, or because truly your sample was unusual, cannot reproduce it. That is interpreted by most as a reproducibility problem, but in fact, that does not mean that your experiment with your sample size was wrong. It just means you shouldn't have generalized. So that's number one. The issue of normalization, the biggest advantage of being able to develop these approaches is that you are able to then leverage a larger sample size because you can do the meta analysis across experiments and measurements that were performed in a large number of sites by a large number of people separated by time in space. But unfortunately, this is extremely difficult due to the batch effects. And even large NIH projects like GTX and ENCODE, HAPMAP even in genotypes have struggled with this and did not come up with entirely successful ways of handling this. We of course have much better methods these days, but they're still not perfect. They still can't integrate over all the variants where we covered the variants. This is, I agree with you that in that sense, there's clear opportunity for additional development. I know that people have not given up on this, but it's an extremely difficult challenge. I just wanted to add, I think Carlos said that IBM, Watson failed not because of the AI, but because of the data. And I think part of this is in the past, we read a lot of genomes, we sequenced a lot of genomes, we sequenced chip-seq, all these different characterizations to annotate the genomes. And while that data is really useful and can be used for predictions, and we need that strong statistical analysis of those sorts of data sets, we also now that we have genome editing, we have amazing ability to synthesize sequences and test different things with massively parallel reporter essays, we need to think about the type of data we really want and what influences we're making at the moment. For example, with all of these attack-seq peaks and all of these epigenetic marks, how many regions are actually functional? And can we learn anything from which ones are functional and which ones aren't? And that needs to be done carefully because if you do that in a single cell line or a few cell lines, you would be making incorrect assumptions. So I don't think that there's a huge disconnect, I think it's looking at sequences and doing genomics with a sort of correlation approach. And now that we're trying to get to causality, we need to think carefully about the true functional assays that we use to integrate. Sure. Sorry, I just want to make one last comment about that, kind of emphasizing maybe a different aspect of it, which is that if you're doing cell culture and you're using a specific serum and then you get a different batch of serum, your cell culture effects may be different, right? And I think this is somewhat of your point as well, is that how do you reproduce the phenotype when the phenotype changes with the serum? And so when thinking about high throughput phenotyping, do we want to focus on phenotypes that are large and reproducible regardless of what the environmental lab conditions are? But then you may not be actually studying phenotypes that are of relevance to the organism, right? Because you may admit those subtle phenotypes that actually change when you get new water and your water's now from one company as opposed to the other because it was cheaper that day. This, but that really matters when thinking about high throughput phenotyping and what phenotypes are gonna be not changing given those differences in conditions and changing and how that relates to what's relevant for the organism. I don't know. I'll just make one last comment. I think in reality, this is the same thing you face in development of pharmaceuticals, right? That essentially the beta testing of the product is in a very small group of individuals in reality. And then all of a sudden when you take it to a population level, you have hundreds of thousands of people taking it and all of a sudden now you start seeing adverse events and it's just you finally get enough statistical power to detect admixture effects. Whereas in the random clinical test with a very actually smaller number of patients, you didn't see that. And I think at the end of the day, we kind of face the same thing in terms of experimental reproducibility. It's just there aren't enough people actually trying to reproduce the same findings, which I think is inherently a problem these days. There was a comment first. Hi, Sushmita Roy from UW Madison. I think my question is related to some of the things that were discussed. So all of you kind of alluded to the importance of a regulatory network and how it connects from the genotype to the phenotype. And Joanne also mentioned about some components of the regulatory network and how it might control the phenotype that you're studying. I was just curious as a community, how should we go about thinking about identifying, changing the regulatory network and how it actually impacts the phenotype? Does it entail collecting systematic gene expression data sets together with attack seek data sets and building on these data sets? What would your thoughts be on that? So I think it, well it depends on your perspective. I think that model organisms can be really powerful for these sorts of questions because we can actually, you could think about the type of organism be it Siona or your favorite organism where you have a large population with high sequence variants or high polymorphism rate. And then we can actually go in, genome edit it and look at a specific phenotype across the population across the different genetic backgrounds. I do think that maybe people will not like me saying this. I do think attack seek and all these epigenetic approaches are super valuable but I don't think they're enough. I think we need to do functional validation or at least validate our predictions that we make based on those data sets. And I think to do that we do need better oligosynthesis technologies so that say you have a taxi, people have recently done this in disease, you find a loci that have open chromatin and you have variants that change the chromatin state that are predicted to cause a disease or a fate change. We can't test those. The region is 1KB. People are testing 100 base per region in reporter assays, they're not the same. So we really need to be finding a way to test these with reporter assays but then do genome editing to see if we can see the phenotype. And I agree with tissue culture assays that's really hard to see. Is this really a phenotypic change that mirrors what we're seeing in an organism? And it would be really great if we could think of ways to couple that to model organisms in a really effective way. I want to draw one distinction here between two of the methods that Emma mentioned. There are reporter assays where you're looking at a sequence in a completely artificial context in a way just with the same trans environment. And then that's a slightly different thing than modifying the actual nucleotide in the genome itself. I think this is a really critical difference. I don't know if you want to expand on that anymore but I think the community really has to recognize that a reporter assay is not quite the same as the activity of an element in context. Yeah, I completely agree. I would say you absolutely need both though because as soon as you start manipulating the genome you change the cell identity and you can't read out what's truly happening because the cell is a different type of cell potentially. So I think you need the reporter assays and you need the CRISPR screens and we're seeing now these really high throughput CRISPR screens which are amazing and we should take advantage of but we have to think about in what context we do them so we can truly read out phenotype. Yeah, I completely agree. Just also want to throw another comment with regards to regulatory regions because those are sometimes some of the most difficult regions to have good sequences and good assemblies and we're seeing a lot of perhaps variation that is important for phenotypes and it's exactly regulatory regions and that's the hardest thing. So I was wondering about, let me Eric could make a comment about the genomic resources, the high quality regionals, those are the particularly difficult to get. A lot of the genomes that are being made available have very poor regulatory regions and because of GCs or structure or whatever, rapid developments and that's a major challenge I think. We're seeing for the genotypes, for phenotypes a lot of the importance of these regulatory regions and I think it's a big challenge that you all face. I want to say one thing in response to Sasmita as well which is that it really depends on your question, right? And I think that that's something that we have to keep in mind is that CRISPR manipulation, if you're trying to get at what is the effect of this gene on phenotype, that's fantastic and I think CRISPR-A and CRISPR-I are also good ways to manipulate expression but if you're interested in, are specific hubs of the regulatory network involved in a phenotype and you have perhaps phenotypes across large taxonomic scales, then you can ask that in a different way using comparative genomics or comparative expression and say are these genes the hubs in all of these phenotypes and if so, then maybe we need to go and manipulate it in an organism that has tools for manipulation but I think that there isn't a clear answer because it really depends on what is the question that you're trying to answer. So there was a comment in the back here. Yeah, Paula Mayby, University of South Dakota. So I just had a comment on this gene-to-phenotype connection that has come out multiple times. So it seems like the speakers throughout the day have really done a terrific job of demonstrating the diversity of phenotypic starting points for comparative genomic studies and all of the different things that can come out of that but it's a bit of an elephant in the room to how we systematize that phenotypic data so that we can synergize our studies. So we can align genomic data and actually there are mechanisms in place, ontologies, to align phenotypic data but there really aren't resources that span all of us that we can, where we can find each other. So as you were saying, Joanna, you could collaborate with people if you knew who they were but if you had a resource where you could find that phenotypic data that would also be a starting point. So I think that part of the elephant in the room is that those data are very sparse. So if you could make a huge matrix right now of all the phenotypic data across all the biodiversity that'll be sequenced with the comparative genomics, for example, it would be unbelievably sparse but it would still be there and we could look at it and we could, for example, strategize using phylogenies and pick up a thread and follow it in a very strategic way looking at sister taxa, ancestral and derived genomes, ancestral and derived phenotypes. So I wondered what you thought about that so there's a difference between, I guess, what we need and mechanistically how we create it. Do you think as a panel that we need that? Would that be helpful? When you think about a phenomic resource how do you think about it? Yeah, it's a tool to build bridges, I think, right? I think it partly starts with being open to communicating and not having judgment about whether you're in human or mouse or drosophila so that we can actually be in the same room, for example. So I was at the genomics of rare disease meeting at the Wellcome Trust last year and I was quite surprised to be invited but I think that it's that, like you say, having the opportunity to meet each other and to talk and in order to do that I think we have to be more open towards model organisms, the use of non-model organisms and to break that prejudice against, oh, but that's not in human and I think that would help a lot in terms of allowing the integration and people to communicate so that you could build that type of resource. Yeah. Yeah. Hi, I don't know if this is the right session to bring this up but I don't think it's really been discussed today and that's comparative genomics of our companion animals. There's about 100 million dogs and 100 million cats in households across the US. Many of them go to the veterinarian, they get vaccinations, they are selectively bred often for phenotypes that have nothing to do with production, they have to do with how nice they are and how cute they look to us and there are also pretty well phenotype disease incidents and genetics and their genomes aren't are fairly well annotated so I don't think we're really making use of all of the different things that could be evaluated with that resource which has maybe a better direct link to the human diseases we're talking about because of the environmental and nutritional and other types of similarities that the species have. I think that's a fantastic point and I wanted to follow up on a related point which is that many of the zoos that work with us actually keep track of their large mammals on a daily basis. They have the most detailed medical records in the world for primates or for primates in zoos but there's a lot of issues in terms of getting that kind of data shared and integrated and it's basically not used at all by our community but there could be some really good opportunities there as well. Just to add to that bandwagon from a sort of policy perspective nobody owns one health, right? So NIH doesn't, right? Purpose of NIH is to improve human health and as we saw with all the work on trying to defund research in chimps and anything that had to do with chimp health it was one way to shut not funding it, right? So as many of us who for example work in canine genetics and just got tired of throwing in grants to NIH that didn't get funded on canines and then just sort of lifted it in other ways I do think there's a massive opportunity to think about those opportunities from a cross agency perspective and for enlightened program officers to come together on sort of catalytic funding that could be set aside for this which is often exactly what you need to do because it will get killed in existing study sections as the powers that we aren't gonna go out on a limb and support that kind of stuff so I really do think it's up to the enlightened feds to come together, especially around one health which I think currently is totally abandoned. It also goes back to the discussion earlier about educating your reviewer colleagues. I think you have to build the right review panels, right? There's just no question around that. You've gotta get maybe some of the people in this room maybe people in other rooms that kinda get it because as many of us know right the key to getting your grant funded is to find the right study section and in this case it just doesn't exist so you have to create it. Yeah, there. Following on that note I do think this term model organism is we should just do away with it. I think that it's model, every organism is a model for a particular question and I believe that it's been co-opted essentially to sort of reflect the hegemony of molecular biology over other kinds of science and so different organisms are suitable for different kinds of questions. Some are really suitable for mapping genes, genotype to phenotype, et cetera but I think we need to be careful in as we explore ways to understand biodiversity and the diversity of phenotypes need to be careful about the sort of continuum of model for molecular biology versus model for ecology or natural variation or whatever it is. So Carlos pushed me over the edge. This is Steve Ellis from NSF again. So I just want to point out the obvious. It's super easy to engineer a mutiny at the funding agencies. Submit more proposals on a given topic and so just do the math. If we have 40 proposals on a panel and an extra 10 show up that all of a sudden are on our coordinated theme of comparative genomics in cattle we have to recruit reviewers with that expertise and that will change the panel discussion so you can do it in the existing systems right now by participating fully and strategically. So I think that's one point to be made and the other I want to tie Joanna's presentation and Tom's together. Your idea of replicated essentially natural experiments or the parallel for me is lots of dairy farms do the same thing over and over again and Tom I wish you had made a bigger deal about the imputed genome and the story of Joe Citizen and Jane Citizen using whole genome selection for literally for fun and profit and that ties together with the companion animals. We're not nearly leveraging the opportunity in the imputed genome resources GWAS and doing that sort of predictive turning the P to G thing on its head. We know the phenotypes. We're getting better information about the genomes and I say genomes then let's drive that into mechanistic detail now. I just wanted to make a quick comment again following on this same line of discussion that I hope in the meetings we have today and tomorrow we can start to maybe put forth some ways to characterize for different species what might make them a good model for a specific molecular function or physiology. Some of those things have been defined and I hope that's where the agencies can help to maybe put that information together. Part of that certainly is about reviewers and reviewers understanding those models but I wonder if there is an opportunities to join forces to help groups define those models as well in terms of funding for what makes a good model. And I can't resist from what Steve just said about livestock genetics in comparison to a lot of the other organisms. There's this incredible wealth of genotype and sequence data and phenomics data and now high throughput phenomics data that's becoming available. I suspect those same things are available with a lot of the companion animals. How do we put the compendium together to see and communicate which ones are useful as several of the other folks have mentioned. This is another comment. I think that's an important distinction that was just made because I mean most of biodiversity can't be manipulated in the laboratory. I think we could think about cell lines and things like that but I think it'd be interesting to talk about what sorts of, how far can we go without crossbreeding without manipulating or culturing in the lab. And one question I had from Joanna's talk was when you brought the fish into the lab to sort of control for environmental effects. I'm curious what we can learn from understanding natural variation in the environment in addition to in the lab. In other words, the expression patterns we see in the lab are not what occurs in nature and yet at some level we need to do those controls in order to sort of see what part is genetic but we're gonna be missing a lot of G by E interactions and of course we're gonna be missing the environmental interactions. So I'm curious, many different dimensions we can think about of studying organisms in nature without looking at the more manipulative aspects and those are certainly challenging in terms of causality but I think they probably can answer questions that can't be answered in the lab specifically. Am I supposed to respond? Yes. No, I completely agree. I mean, I think all of the original work was done in the field. I mean, we didn't really have any money so we were limited to that and the question was well, if we only have a limited amount of money and we're interested in how these may be adapted to hydrogen sulfide, first thing we did was go out and sequence RNA from wild caught individuals and that gave us huge insight into what was going on. All the detoxification, the fact that we didn't see changes in endogenous hydrogen sulfide genes and we had also hypothesized stress response genes would be like this is where they live. There were no stress response genes going on. So we learned a lot from the wild study and also from that we identified SNPs from the RNA-seq data and started actually identifying some of the possible mutations that were different between sulfitic and non-sulfitic individuals just from the wild caught samples. So that, I mean, to me I think that looking at things in the wild has huge value and a lot of the things that I'm interested in like Antarctic fish, I don't have the means to study them in the lab and so all of my questions are limited to when we can sample them, how, you know, like when we're getting them in the year, where we're getting them from and they're not gonna be manipulated in the lab but I don't think they're any less valuable because of that. And that probably means we have to do a better job at describing our environments, you know, more quantitatively and in more detail so we can compare across study. So just a couple more comments and then we'll wrap it up. Yeah, so I'm Matt Hufford from Iowa State University and I'm a MACE Genome Assist and just one perspective in terms of connecting, you know, the genotype to phenotype, you know, that we have with plants. I mean, we're getting to the stage in MACE now where we have tens of genomes that we've assembled to reference quality and people have been very, they've thought a lot about which lines those are gonna be based on, you know, the phenotypes that are interesting to the community and so we've been very specific in the lines that we've chosen but simultaneously as we've been doing the assembly and annotation, we've also been developing the resources, the mapping resources, the CRISPR knockouts for those lines that are being assembled so that, you know, once those sequences are released to the community, there's also all these resources that can take a mapping population out to a number of different environments, measure all these different phenotypes so I think it's not just about delivering a sequence, we have to be thinking about, you know, how can we enable a community by producing all these supporting resources as well? One more for Merez and then we'll wrap it up. So I just wanted to pick up on a couple of threads. So one was regarding prioritization and the efforts that people would make to establish model organisms. I think one of the things that would be terrific is to have, you know, truly democratized the tools of comparative genomics and genome assembly such that people just have a box in their lab, you put a sample in the box, it spits out a reference genome. And I wanna highlight that I think actually, to some extent, what we should think about doing is not so much how do we prioritize reference genome assembly, but is it possible to replace DNA resequencing with, you know, some form of genome assembly such that whenever you would do DNA sequencing, you just instead put it in the box and get a reference out or get an assembly out. Now that's the kind of big thinking we're asking for. 10 years from now, that's the dream. Thank you, everybody. We need to wrap up because AV folks need to go home. If you are, have reason to believe that one of the agencies is reimbursing your travel, you can be here at 8.45 tomorrow morning to find out how that works. Otherwise, we'll see you at nine. Thank you.