 Okay, great, thank you very much for giving me some time to speak to you about some things that we are working on. I'm going to take off from the point of secondary findings and not likely for the reasons you might expect, but I think we fell into this as part of our program and it taught us things that we were surprised to learn and I think are actually useful and helpful for us thinking about how we want to go forward in this realm of research. A couple of minor distinctions here, the distinction of secondary findings from incidental findings is an important one, the former being an intentional search for actionable mutations, much like several of the projects that Gayle described. And the key thing here is that the search for those is independent of the indication for why the sequencing was done in the first place. And those are distinct from incidental which are sort of arbitrary inadvertent discoveries of variants stumbling upon them, et cetera. And suffice it to say, I can summarize a fair amount of angst in a short sentence to say these have been fairly controversial, much derided and often actively avoided part of genomics research and I'm thinking we need to change that because in fact secondary findings are predictive genomic medicine. I would suggest that this is exactly what we want to be doing is going out and finding disease susceptibility in situations where we otherwise might not have even had a reason to ask the question if it was there or not. All right, so precision medicine, predictive medicine, genomic medicine is part of precision medicine. This notion of tailoring risks to individual risk, tailing treatments to individual risk profiles, stratifying risk of disease, stratifying natural history, variation complications, prognosis, treatment responses, et cetera, maximizing efficacy of interventions, minimizing off-target or side effects. And I would say that what this is all about is shifting from a phenotype first approach to a genome first approach and using the genome to tell us what we should be doing, both in the research realm as well as ultimately in the clinical context. And what that will start to look like is having genomic data first and then following that with iterative post hoc phenotyping that is driven by the sequence. It is not driven by the concerns or the questions of the patient or the clinician. That's a bit radical. It makes some people very uncomfortable, but I think this is a activity that we have to do. We sort of stumbled on this originally back a few years ago when we were working with a wonderful collaborator at our institute, Chuck Venditti, who works on typical pediatric onset organic acidemias, combined malonic and methyl malonic acidemia. We set out to figure out what the cause of this was. And this was at the time what we were studying was an autosomal recessive disorder that involved childhood organic acidosis, severe metabolic decompensation, CNS infarction, irreversible stroke, coma, and sometimes death. This is a fairly typical severe pediatric onset metabolic acidosis disorder. So we sequenced a trio, followed it up with validation, Chuck did beautiful functional work and found that it was caused by mutations in a gene called ACSF3. So that's now the very well-established old-fashioned paradigm of how you find a mutation that causes a rare disease. We actually took another step there, which is we wanted to know because this is a subset of a metabolic acidosis phenotype, how common could this disease actually be? So we went into ClinSeq data. And ClinSeq data, as I mentioned, we started in 2006, had about 1,000 samples of people that we had exome data on. So we just went and asked a simple question, what is the carrier rate of that disease and that population from which we could back calculate the frequency of the disease? And so sifting through the ClinSeq data, we came upon a patient who was homozygous for one of the variants that one of the kids had who had the severe metabolic acidosis decomposition and stroke and said, well, that's probably like those variants that Gail mentioned in HGMD, lots of people, including probably us, have made mistakes and incorrectly attributed causation to variants that are in fact benign and were coincidentally present in people with the disease. But in ClinSeq, we have the opportunity to go back and re-phenotype people. So we have the opportunity to go back and ask a question that we never thought to ask when we put the study together, which is, could a patient in ClinSeq have any attributes of combined malonic and methyl malonic acidemia? So we brought the patient in and did testing on her and showed that she had at the age of 66 years old, about 100 times the upper limit of normal of methyl malonic acid in her plasma, about 100 times the upper limit of malonic acid. And then when we scanned her, she had multiple scattered small infarcts that are imaging signal-wise very similar to the strokes that the kids get when they decompensate in the nursery. And then we interviewed her and asked her more questions about her history. Turns out she had a lot of unexplained late onset neurological symptoms. So what did we learn? What we learned is that our pre-hoc conception of what we thought that disease was, was completely wrong. We thought it was just a pediatric onset, severe metabolic decompensation, stroke, and death disease. But it is, in fact, more than that. It is a disease that can present throughout the live spectrum. And we had just never before been aware of that. And no one thought to ask that question until we looked at the patient which was triggered by the genome variants. And so what this told us is we have the ability to do a genome-first ascertainment approach to research that allows us to get beyond some of our ascertainment biases and come to a full understanding of what the genotype-phenotype spectrum is, which is, of course, the essential prerequisite to understanding genotype-phenotype correlations. You have to know what the spectrum of disease is. We've extended that through a number of other genes, collaborating with various groups to look and to identify rare forms of late onset diabetes and cardiac symptoms. And then the secondary findings work that we started, which I would suggest actually opened this field originally by screening our exomes. Back when we only had 500 in some exomes, individuals ascertained for possible cardiovascular disease. We asked the question, what fraction of them have inherited susceptibility to cancer should be unrelated to why they were ascertained for cardiac disease? And in fact, we found a substantial frequency of that. We followed that up with papers on both cardiac dysrhythmias and hypertrophic and dilated cardiomyopathy. And then the most recent one is screening for malignant hyperthermia, classic drug-gene interaction phenotype inhibits. And Gail mentioned this paper and validating this in a broad way across the genome, with the exception that one of the things that we did in these studies is we actually went back and re-phenotyped the individuals for the traits. And this is limited in the ability to do that. But then we wanted to take an even larger step away from phenotype. And again, let the genome tell us what we should be looking at. And so what we did is we said, let's look across the genome for all null variants that our participants might have. This was done in a general way by the 1,000 Genomes project previously. But again, we wanted to couple it to phenotyping. So we searched the exomes for all null variants, then filtered that for genes that are known or have there's good evidence for the gene function being mediated by haploinsufficiency as its mechanism of disease, limited it to things that are plausibly related to disease and not just benign traits. And then whatever variant the participant had, we would go back and phenotype for them irrespective of what it was. And I will tell you the important thing about this is nearly all of the traits that we phenotyped them for. We had no pre-hoc data. And we had to create that post hoc. So that's what this post hoc phenotyping is about. What our result was showed that we have about 50% yield of disease. That is, for any participant who has a loss of function mutations in a gene that appears to have that as a mechanism of disease, half the time they had the phenotype and half the time they did not. The half who did comprise between 2% and 3% of our cohort, which is a number that is much, much higher than I predicted would have been there. And most of those and individuals were not previously diagnosed with these diseases. And I would have to say, if anything, this is probably an underestimate because for sure we don't know all the genes who act in this way. This is obviously a subset of mutations. We're ignoring missense mutations, which are arguably harder to interpret. But there's many more of them there. Our phenotyping was pretty conservative, as well as the fact that we are sequencing adults here. Everyone we sequenced was over the age of 45. And as you all know, loss of function mutations that cause severe disorders often have severe morbidity and mortality in childhood that would not allow them to survive to the age of 45 to be in a research study such as ours. So what have we learned from this? The hypothesis testing model for research absolutely positively works is a fantastic paradigm for approaching scientific questions and should not in any way be limited or supplanted by what we're proposing here. But I would suggest that it can be complemented by what we're proposing to do. And that hypothesis generating research using iterative post hoc phenotyping of patients is essential for us to understand the predictive power of genomics for precision or predictive medicine going forward. We have to have an unbiased ascertainment of phenotypes in order to understand the full spectrum of genotype-phenotype relationships. And this concept of genomic secondary findings was established by this work. The rate we estimated back in about three or four years ago between 2% to 4% has been borne out by several subsequent studies. I would say that what we're thinking about focusing on is that we need to expand and scale this concept of hypothesis generating research. Because I think this is where the payoff will be for us. Genomics has completely revolutionized how we think about basic scientific research. And it has been slower to change how we think about clinical research. Yet I think there is an opportunity there to do that. How we will do that is not a small task. It is harder to change clinicians' minds about how things are than it is basic scientists. They're a conservative bunch. What do we need to do that? We need cohorts that are thoroughly consented for this style and approach to research. They have to be very highly motivated. Some of our participants, it's actually amazing what they will consent to allow us to do to answer questions that may have no relevance to them. It's quite startling. And we need creative and motivated investigators who are willing to do this kind of work to break down the paradigms that we have and move forward. We can envision this going forward in two ways. We're thinking about putting together a combined super cohort within the NIH Intramural Program of adding variants from multiple investigators, cohorts into a pooled bio resource that would consist of volunteers, samples, and genotypes. And then opening that up for investigators basic scientists to go through those data, look for variants and genes of interest to them, request samples, and then get samples from these volunteers to do their basic science research studies on, again, genomically ascertaining the variants that they want to study. And you can also envision, obviously, as I explained how we could do a clinical translational version of this, which is to have investigators queering that database, make study requests, and then have secondary studies performed and those participants, and then feed those results back to investigators so that they can have their questions answered in this hypothesis-generating approach. We also need to push this into clinical space. So I think we have plenty of data, Gail outlined a lot of data that shows good results and validity of findings and approaches that these approaches can be potentially fruitful today in clinical practice in some specific circumstances. We're pushing that out into the NIH Clinical Center, where we are starting a two-track program where we provide sequencing research resources to investigators. And at the same time, use those same data to mine them for secondary actionable clinical findings and return those to research participants so that these researchers can be free to focus on the research that they want to do and they have others take care of this notion of the secondary findings and get this practice and mindset out into the clinical community because, again, it's hard to change how people do things. So my conclusion is I think we need to think genomically in clinical research and that's a much, much easier thing to do than it is to say because, again, these paradigms are hard to change. Hypothesis-generating research I would suggest is a valid, useful, and complementary approach to hypothesis-testing research and I think if we're going to take genomics to the next stage, this is where I think the excitement is, where the opportunities are, and how we can begin to answer some really exciting new questions and make precision medicine actually work. I'll stop there. Thank you for your attention.