 Okay, so hopefully we don't have to spend a lot of time with this group talking about what the classifications are for variants, but the topic here is variant classification, consistency of variant classification across labs. And just a reminder that the goal for these is to get them into five bins from benign to pathogenic and that we're trying to clean up our nomenclature a little by not using mutation as a synonym for pathogenic because a mutation is any change, not just a pathogenic change and not use a polymorphism for benign because polymorphisms indeed can be pathogenic. This is data from the Partners Lab that was in the very nice ClinGen marker paper that Heidi was the first author of recently and it just highlights that in this clinical lab, the blue bar here is variants of uncertain significance and 71 percent of their results were variants of uncertain significance and these are particularly challenging and many of them are novel variants and we don't know what they mean and that's an obstacle for clinical genetics because you can't take an action when you hit a VUS. A second obstacle is that when there actually is data to inform our pathogenicity classification, we don't agree all the time. So this is a CSER consortium across six labs, Bake Off, we called it, and we picked six variants and asked each lab to classify them and I think you can see there was only one variant that on the extreme left that everyone agreed on and that was a truncation and a well-known gene, so we would hope we all agree on that. But otherwise there was a little bit of disagreement on every other variant. Now sometimes it was just one class away from pathogenic to likely pathogenic from VUS to likely benign, but still disagreement for the other five variants. We moved forward with this in Exome Variant Server annotation paper where we looked for all the incidental findings as another CSER product in the 6,500 people in that server and we found a couple percent rate of incidental findings, but as we're doing the QC, every variant that we wanted to call pathogenic or likely pathogenic, we had recalled by a second observer and so you can see in the figure there were 79 of those and when they were recalled blindly, only 35 were concordant. Sometimes they were one class away, sometimes not, but there was a high discordance rate. We had an independent, another blinded reviewer go through all of those and took the consensus 2 out of 3 and sometimes 3 out of 4 view and for the discordant variants, 95% of them really were VUSs but they had been called pathogenic or likely pathogenic and that just shows a bias towards over calling the variants and that's particularly problematic in medicine. It's a mistake either way to call something pathogenic when it's not, leads to the wrong care for the patient, to call it a VUS if you can call it pathogenic, leads to loss of ability to do interventions for the patient. I did want to make the point that when we got through this whole exercise and compared our final list to other experts in the field, there was 99% concordance between the calls of our pathogenic and likely pathogenic so there is hope out there. We can do this but it takes more work than you might imagine. So the ACMG recently published this paper with Sue Richards and Heidi Rehm as the senior authors on guidelines to try and standardize variant interpretation so that we're all doing it the same way. The system is complicated so you're not meant to read this slide, this is just to give you the impression of this is, the system is complicated and so each one of these, we don't have pointer control here, each one of these boxes has population data on the side and level of evidence and for each box there's little codes BS1 or PS1, B for more benign, P for more pathogenic and you take all these lines of evidence and you develop these little codes and then you take the codes, oh it's a pointer, keychain, wow, wow, that's amazing, technology. I thought you might have had a cat, Heidi, so anyway what you can't see very well on this slide is that there are some of these that are very soft and squishy. This one says multiple lines of computational evidence. This one says co-segregation but it doesn't define co-segregation. This one says well-established functional studies so some of these are a little bit soft in how you might interpret them. So once you have those then there's this long menu of you combine the codes to get to the different classes. I don't know why this isn't such bad focus there, anyway. So the point is it's complicated, hope I've made that point. So what we did, and CSER as we did, another bake off where we picked 99 germline variants and nine of those variants were classified by all nine labs of the U01 projects in CSER and then 90 were classified, it'll eventually get to three but some of them are still at two sites. And each lab was asked to do it by their own lab rules the way they would normally annotate and to apply the ACMG rules. So this is the intra-laboratory within the same lab comparing the laboratory classification and the ACMG classification. So on the diagonal is when the labs own classification and the ACMG match and there was 73% discordance. There were 9% of the cases where ACMG was less pathogenic, 20% it was more pathogenic, but a lot of those more pathogenics were moving from benign or likely benign to the U.S. So okay I didn't do that. Okay well I have a feeling not everyone is on the WebEx, but everyone can get off the New York Times and move to the WebEx. Okay here we go. So if they were discordant the ACMG was less certain more toward the U.S. in about 80% of the time. So then looking at all 99 variants and just to keep you on your toes I've switched the axis ACMG class up here and lab class down here, 80% concordance for the 99 variants across all the reads, again about 8% of the time ACMG less pathogenic, 12% more pathogenic but again ACMG 67% of the time is sort of more conservative, more in the middle. But that's not necessarily a good thing because if you can call something benign or likely benign instead of a V.U.S. if you have confidence for that then that's useful to know. So the ACMG has this category a standalone in order for an allele frequency to stand alone give you a benign classification it has to be 5%. And the labs were saying we don't need 5% if the disease is 1 in 100,000 then an allele frequency of 4% we will call benign. So they're telling us a little bit about what they don't like about the ACMG's criteria. All right, so then the next thing we did was to compare the concordance across the nine labs using the ACMG criteria. And what we found was that about a third of the time all the labs agreed. So that's great a third of the time that's better than it had been a year ago. The difference was only one level so from pathogenic to likely pathogenic or from benign to likely benign one level for about another third. But then there's a group where the difference is among the range in the lab is quite wide and so for this one for example there's a range from benign to pathogenic depending on the lab. So the lab's own criteria is in the light blue ACMG criteria in the dark blue and what I think you can see is the ACMG criteria didn't actually perform any better, they didn't perform any worse but didn't perform any better than when the labs were using their own criteria. But we did better than last year so that was good. So this is an example of just one variant where there was high disagreement and each of these columns here are particular codes. This is the computational evidence code, the functional evidence code, autosomal recessive in trans, the co-segregation code and you can see just by looking at the boxes even without knowing what they are that people did not use the same classification. So this is an autosomal recessive disease for spastic paraplegia. It's exceedingly rare, one in 50,000 maybe but it's a protein substitution, it did have conservation and people applied very different classifications and what you can see is the people who called it lesbenine here used this minor allele frequency as just too high classification. That was really convincing to them. The people who called it pathogenic ignored the high minor allele frequency even for recessive disease and indeed the incisor, the lab doing genomes found this variant in three of 50 people and it did Sanger confirm. So that suggests first of all that the exact frequency of .4% might, allele frequency might have been lower and maybe it's not captured in all those exomes but it tells you even with a lot of data you can still have trouble classifying a variant and the labs are still doing it differently and we need to get better at that. So we did some power calculations, Brian Scherz did this and just to get an idea of what kind of sample size you need under different conditions to actually determine the pathogenicity of an allele and this is if you had a case control cohort, so even with a relative risk of disease of 12 and a minor allele frequency of .01% which we're going to see a lot of rare alleles, you need 6500 cases and 6500 controls to capture that variant's pathogenicity and conclude that it's pathogenic which means if you have a population based sample you're going to have a very hard time getting that data which is why and I appreciate Jeff referring to this some of us think that as much as ClinVar is an extraordinarily important and useful tool that we're going to actually need more data to bear on this problem and have considered the idea of a broader data commons where the clinical companies that are not research based are required to share data and not just one or two variants per person but a significant amount of the genomic data should be stored. So I'm going to thank all the people listed on this slide which you can read later on the web and pass to Heidi. So you can just advance the slides and I'll talk from here. Thank you, Gail. So I wanted to talk a little bit about ClinGen really to say, you know, as we look at all these differences in variant classification how are we going to make it better and that is really the intent of ClinGen is to try to come together as a community and standardize how we approach this and involve expert consensus in the classification of variants. So this is an infographic that we've developed for ClinGen to really point out that little laser pointer. Thank you. The key questions we're trying to ask, you know, first starting with sharing data across all the different stakeholders and asking is this gene associated with disease is a variant pathogenic information actionable and then really building a curated knowledge base to support patient care. Next slide. So one of the first databases as part of the ClinGen project that we focused on is really the ClinVar database is a place to share data. I will say even before ClinGen was funded, you know, two years prior working with NCBI, there was a lot of discussion about what kind of database we should build. The first interest was building a clinical grade variant database where the only thing in it would be variants that we fully understand and can be used clinically. And then we all said, yeah, but there'll be nothing in it. And so we and even if we can put stuff in it, it will take a long time to get that information in there. So we did make the decision to create ClinVar in a way that allows everyone in the community to share information at whatever stage it's at, because that is how we operate in clinical diagnostics today is we read every last research paper, go to every research database and that's what we do to do it today. And so we decided let's allow that to happen. And so indeed ClinVar takes data from every source and allows submitters to submit that data into it. We link up to local specific databases to get data in there, but still link out to more detailed information in those databases like FarmGKB, for example. This just shows you the stats on what is in ClinVar today at the bottom. Next slide. So one challenge is when you have all different sources of information, some of which are high quality and expert consensus and some of which are not, you do need a way to allow the community some transparency into the level of review. And so we've developed this star system where ClinGen actually approves groups to be in this three and four star top tier and you can see the groups that have so far been approved there whereas the majority data in ClinVar today actually exists in the lower tiers where they are single submitters. Most of the data today coming from clinical laboratories that have shared their own assertions. We did recently decide to create two levels within single submitters and this will be launching in June where those groups that provide their criteria for variant assessment and attest to a comprehensive review will get a single star and those that don't won't get a star and these assertions will override those. So that's being launched in mid June to again create some more differences in how these things are being done. Next slide. Now if you look at ClinVar today there are over 118,000 variants that have been interpreted and about 11% of them have more than two submitters that allows us to compare those assertions and what we see is about 17% are interpreted differently within that data set. So again it emphasized the point if we look broadly across thousands of variants that there are differences in how we approach this. Next slide. That being said if you work together you can resolve them and this is just a slide that shows an exercise three of our laboratories went through after our first submissions where we took a hundred differences and we worked to systematically resolve them. The first 75% were simply resolved by sharing the actual evidence we were using. The rest were resolved by reconciling rule differences in how we were classifying variants. The last variant required expert input on a functional assay but all of them were resolved. So by working together, coming together as groups and sharing experience and evidence these can be resolved but it will take time. Next slide. So we at the bottom of this figure that I showed before we are working to develop through ClinGen a curation interface that will help allow a common place for groups to come together both resolve differences as well as do expert consensus on the variants and you can see down here the ClinGen work groups that have been created that are will be able to operate in this space. Next slide. We lost it again. So I'll describe the next slide in advance if you've seen it or if you're on the WebEx you can see it. So in this curation interface we will take the ACMG rules and build off of them creating more consistency. So I took the rule set and I highlighted in blue those rules where a tool or resource supporting the rule would make it much more systematic to apply as well as some that quantitative approaches that we can add will help improve the consistency. So hopefully we can improve how we're doing this by a common interface. Next slide. And then the other thing is that each of our clinical domain working groups will help apply more specific rules to classification so these are some of the charges of the ClinGen clinical domain working groups is to define both genes variants and rules for classification. Next slide. Because I think one thing we haven't talked about is the fact that you really can't interpret a variant in the gene if you don't even know that the gene is involved in disease. So we also have a focus on interpreting genes and time is running low. So I'll just skip to the next slide. This is the grid of sort of gene assignment for strength of evidence that we've developed and Jonathan and Christa Martin have co-chaired this gene curation committee to develop these rules. And the one last slide I have is to show how we've started to apply these rules in different clinical domains so you can see in the first column for our work share and plan helps oversee our clinical domain working group in hereditary cancer. They apply this to pheochromosatoma and paraganglioma genes and we're able to show that almost half of those genes don't have sufficient evidence for implication in disease. We did the same thing in hearing loss. Similar a large bucket of hearing loss genes don't have sufficient evidence. We've now applied this to 1,504 genes in the baby seat end site project that looks a little better although I will say we started with the most well-established genes and that's why it looks good so far but that bottom bracket is increasing over time. So it is true that we all have to think about gene classification in addition to variant classification. Next slide. And that, this is my last slide to sort of show how we might use these rules in the clinical space to define what genes should be included and what type of test. For example, only those genes that are definitive or strong evidence for involvement in a predictive test or a secondary findings return versus moderate evidence genes could be included in diagnostic panels but really leaving the limited or disputed evidence genes for only when you're doing full exome and genome analyses where you're already going into the unknown. So that's just a rubric that we're thinking about in terms of how to take these rules and use them clinically. And the last slide is just acknowledging a large number of people that have involved in ClinGen. Turn it over to Dan. It doesn't stretch. It doesn't stretch any. Maybe we should sit in the change please. So Teri and Howard, we have some summary slides which we were planning to show at the end of the discussion. We can show them now. Well, or do you want to turn them into questions for to generate discussion? I defer to the chairs. Pardon me? I defer to the chairs.