 Okay, so great. For the next section, we have the Interpreting Variance and Actionability Group with Deborah Leonard, Levi Garroway, Gail Jarvik, and Elaine Lyon. So why don't we go ahead and start with Gail Jarvik, please? Is there something we can do about the slide fitting? Yeah, let's see if this continues to be a problem. Okay. Maybe it'll work. Thank you. So hopefully I don't have to convince most of us here that in order to really be effective and safe with genomic medicine, we need to know what a lot of variants do that we don't know now. Germline variant pathogenicity is a critical problem. We need to interpret these, and we also have to agree on a consistent way to interpret these variants. I think we're all aware that every person has many, many rare variants. When we look at a genome or an exome, we see very many rare variants, and we need to know what a lot of these things do. When we look in ClinVar, which is a very important tool in clinical medicine, there are only over 100,000 variants that have clinical assertions for pathogenicity. And per Heidi Rehm, if there are multiple assertions for a certain variant, 17 percent of the time in ClinVar they do not match. So you've heard a little bit about the CSER bake-off. This is the bake-off 2.0. We had another one a year ago that I won't be talking about. I mean, I'm going to highlight some things that Katrina didn't go into in detail for the bake-off. So we took 99 variants, all nine labs looked at, the rest of the variants, three or more labs looked at, and we looked for agreement. And the first thing is we classified them using our usual lab classification, and then we tried to use the brand new ACMG criteria to classify them. So this is actually matching within a lab. So the same lab using their own versus the ACMG criteria. And so there's about 80 percent concordance, so that's good. But one of the things you can see is that up here, these 11 variants are variants where the labs were very comfortable calling them benign or likely benign, but the ACMG criteria were not. And so that's a trend that we saw, that the labs wanted to call more things benign than the ACMG criteria, let you call, and that has to do with the rules around making something benign and ACMG versus what the labs use. So this is sort of an iterative process to inform back the ACMG guidelines that maybe they're too conservative about when they call something benign. And similarly, they're a little conservative on the other end as well. We saw this slide as well, which is the agreement across labs a third of the time. We all agree a little bit more of the time we're off by one level. But what I want to highlight on this is that this is using the ACMG criteria or the lab's own criteria. And so you can see it doesn't take a statistician to see that using the ACMG criteria did not improve our agreement across labs. And so that was a very interesting finding and a little bit concerning that our consistency didn't improve using these guidelines because that was a major point of them. The ACMG guidelines allow you to think about the variance in a standardized way, but also that should lead to more standardized conclusions. When we dig into specific variance, and this is where I hear Jim Evans telling me to stay out of the weeds, so you're going to just stop me, Jim, when I go too far, that this is just one example and it gives you some idea of the things that we're learning about the rules. So this is a variant, the variant itself isn't that important, but you can see that under each laboratory's own rules, this is how they would classify it, and under ACMG rules, this is how they would classify it. So there's correlation. Once people have kind of made up their minds, sometimes they find the rules to follow their hearts. Also you're allowed to overrule the ACMG classifications, that's one of the things that's allowed. So sometimes there's an overrule here, but you can see that for the labs that wanted to call it pathogenic, they were much more interested in the co-segregation evidence. They were much more convinced by the co-segregation evidence. So one of the shortcomings of the criteria that we found is that co-segregation is undefined. It says to do these co-segregate, but there's no definition around that. So people will use that criteria differently. Functional evidence has been another problem area that we've found across many, many variants. Functional evidence is how does it behave in a test tube? But it's supposed to be a well-established criteria that correlates with disease. And there are actually very few of those, with malignant hyperthermia maybe being a very good example of one that we do believe. So many times people are using these functional evidence that are published without really meeting the standard of being well-established to other people. Computational, this is your group score, your CAD score, your polyphen, and people use that more consistently. There was high scores for this. A big difference here is minor allele frequency above the disease frequency. You can see the people who noticed the minor allele frequency was too high for the disease were much more likely to not call it pathogenic. In fact, this variant was quite common. What was interesting, it's a recessive disorder and a few of the labs said, well, we don't apply that criteria for recessives, but it's pretty simple math and should be applied there. You could see a couple outliers for a missense gene where missenses are rare in the gene. That's another one where there's no quantitative criteria. What does that mean? Does that mean below the mean, below the fifth percentile for having missenses? No definition. We also found that the phenotype specific to the gene in this case is really an error because there was no phenotype in this case, but this was very misused in many, many classifications. It really meant there's really only one gene that does this, but it was often invoked for well-known genes but for highly genetically heterogeneous conditions, which was not in the intent. We're going through, looping back, trying to find ways to clarify the guidelines, finding common errors. There were a few common errors that were made in interpretation of the guideline or application of the guidelines, and we are presenting this at the American site of human genetics. We have a four-hour educational session at the American College of Medical Genetics meeting, and of course this will be written as a paper as well. We came to consensus after many calls on 69 of the 99 variants, and the rest we just don't agree on. All right. I think we would all like a sorting hat for variants that would be great, something magical that would be really helpful, but there's no question that there is a lot of more work to do in getting variant classification more consistent across groups and improving the guidelines. We want to continue to identify the sources of variation and how we classify variants, and also, as was pointed out, CSER has been a major depositor to ClinVar, and we feel like that's an important part of our mission. Of course, if variant classification is wrong, then patients get managed incorrectly, so it's a big clinical problem, and I can say as a clinical geneticist that it's common for me to disagree with the variant classification that a clinical lab has provided for one of my patients, particularly likely pathogenic. I often don't agree with them that there's not much data there. We are very interested in expanding to non-CSER labs, and in fact, non-academic labs, companies. Low-penetrant variants are a particular problem. You may have noticed, Katrina said we had 98 variants, and I said we had 99. One of them was dropped out of some of the analyses because it was a low-penetrant variant. There is no classification for that under the current ACMG system. You would call it benign for high-penetrants, but some labs, they said, well, it's pathogenic for low-penetrants, I'm not going to call it benign for high-penetrants, and just didn't classify it at all, so that's a problem that's entirely untouched. So what are our opportunities to maximize variants that are classified? Well, first of all, I'd like to stress that I think that exome and genome-level data is the right data for this problem. It allows you, obviously, a broader look at the whole genome. It allows you to come back to new genes that are associated with your condition in the future, and it allows you to take variants of interest from the public. So I know that the exome variant server, for example, gets many queries. We see this variant in your server. Can you phenotype the patient for X? And the answer to that for the exome variant server is no. But for CSER, the answer to that can be yes. We could come back due to an outside request and say, yeah, sure, we'll bring that patient back in, and we will look at them carefully for that phenotype. Similarly, we want to re-analyze the sequence data when there are phenotype changes in the patient, or when there are new genes that are associated with the conditions of interest. And then another opportunity for variant classification is that there are multiple initiatives to do really high throughput assays to classify variants for pathogenicity in an in vitro way, and we could partner with those activities in order to see if a variant that we see as pathogenic or not pathogenic that the consistent result is found in those high throughput assays. I just wanted to dig in a little bit deeper on the re-phenotype idea and give you a specific case. So this is a patient who actually was collected through a colorectal cancer phenotype. But as we went through the whole exome, we found this variant. And this variant is published in the literature associated with this disease, hereditary hemorrhagic t-line dictation. If you see down here, these little red dots on the person's lip are these vascular anomalies. And so you see these on the patient's skin. You see them on their tongue, their mucus membranes. They get nose bleeds, they get GI bleeds, and they have larger arteriovenous malformations in their organs, and these can lead to hemorrhaging in the lungs and, in particular, hemorrhaging in the brain. So it's a very significant disease, and the disease frequency about 1 in 5,000. This variant's frequency is about 1 in 5,000. So you're already a little dubious of this report in the literature because there are many known variants for this disease and this gene, and when one variant starts to account for the whole group, you're a little concerned about that. And then when we were able to bring this patient back, and on return the patient had a completely normal skin and mucus membrane exam, they had no history of nose bleeds whatsoever. The patient was 56, and this patient, we don't even really need to CT them or image them for ABMs, they're unaffected. And so we can put this data to ClinVar, and the next person who comes along with this variant, particularly if it's a young child, you'll have better data about that this clearly isn't a highly penetrant variant if it's a disease variant at all. So this is a real opportunity to improve medical care for people in the future. Other opportunities, we talked about the tumor sequence data and how people are finding actionable results in tumor sequence. That has to be followed up downstream to see if that changes outcomes for those people. The electronic health record integration, I think, is an important point. It was a good question earlier how it differs from eMERGE, and I would say that eMERGE isn't dealing with exome and genome data, and even in eMERGE 3, they'll have 100 gene sequenced, and so how you get a whole genome or exome captured in electronic health record in a useful way, I think, is a problem that CSER is uniquely addressing. There is a big need for training, clinical training, and training of a diverse workforce. We want to get more patients that are diverse, but part of that is having a workforce that's diverse. We have over 200 clinical practitioners in CSER, which is very unique. One of the reasons we've had such a focus on clinical implementation. Finally, we're very interested in, as Dan Rodin pointed out, the interactions between the clinician, the genetic counselor, and the patient. What's the best way to communicate this information in person? What's the best way to communicate it in reports? Interactions with the primary care provider. And we have an opportunity to look at more diverse populations as well. So gene lists, I think we all know that there's times when a list is useful, like the naughty and nice list, but for genes, gene lists have never really been studied, so most labs that look at specific phenotypes, they'll do a whole genome or exome, but they will look at a list of genes that they know are associated with that condition. Well, that focuses your attention, it saves you time, but it also limits your ability for discovery of variants that are outside that space that may be associated with your phenotype. So this is an opportunity to really study whether that approach is the best approach. For incidental findings, the incidental finding list in general, there's general agreement that an actionable list is useful to look at on a genomic or exomic test, some people would have a broader or a more narrow list, and a list of lists or consensus list may be helpful. So the major themes then today, determination of pathogenicity and sources of variability and interpretation of variants, I think it is a very unique space that CSER has addressed in partnership with the ACMG criteria. We have had the ability to really test those guidelines and see how well they're working for us and see where improvement can be made. Reanalysis of data in light of new information, what are the implications of using or not using a gene list, and results reporting to physician and patient. And I will just end by acknowledging both the bake off team, which I cannot stress how many hours. We've spent like a half hour to 45 minutes on a single variant on some of these calls. So this team has really worked hard. And then of course the whole CSER team represented by the U01 PIs here. So with that, I think we're gonna go to Elaine and then take questions later.