Thank you for coming to this module G. The supplemental module for the last unit of this course. And in this module what I want to talk about is genetic prediction. And let me tell you why I think it's worth doing a supplemental module on genetic prediction. This is a slide I gave you earlier in this unit, and I said there are three factors that'll be important to the success of genomic medicine, or sometimes now called precision medicine. Genetic prediction, individualized medicine, and then finally genetic counseling. I'm going to focus here on genetic prediction. It's something that actually since the first time I did this course, did the recordings for this course, there's been a lot of debate in the genetics literature. As to just how successful we are going to be in being able to predict disease risk from genetic information. We're coming to a time, and we've talked about this already, where many, many people are going to have their genome sequenced. But geneticists are debating just how informative those genetic sequences will be. So I'd like to, I don't think the debate is fully resolved at this point. But I think there's some very important and interesting issues that could be discussed around this notion of genetic prediction. And so I'd like to, to touch on them here in this supplemental module. And, really the issue comes down to this. This is also a slide that I gave you earlier in the course. This is back, I think it was in unit 5 when we talked about schizophrenia. The missing heritability problem. And here I've only highlighted, really, the two right columns here. The biometric heritability, these are not behavioral traits. I guess BMI, obesity is a behavioral trait, but otherwise they're, they're not really behavioral traits. The how heritable we think these diseases are based on twin family studies versus how much genomide association studies have been able to account for. Given the snips that they've identified in the GWA studies. And you can see that, for the most part, most of the heritability is missing. That is, the GWAS accounts for a very small percentage of the overall heritability of these particular disorders. Again, these are mostly non-behavioral disorders, and if anything, this is from a recent publication by Naomi Wray, who's a prominent Australian geneticist. Where she gives the percentage of heritability accounted for by GWAS findings for behavioral outcomes. You can see that they're all very, very low. Less than 2% for bipolar depression, ADHD and autism. About 7% for schizophrenia. Maybe a little bit higher for Alzheimer Disease, almost 20% there. But right at this particular point in time, given what we're finding with GWAS, it doesn't seem we are counting for much of the heritability of these disorders. It's exciting to find these things, but what are the clinical implications? Does this really help us predict who will or will not have the disorder? Does it give us any genetic prediction? Well, to understand the extent to which it gives us genetic prediction I need to introduce a statistic. I'm not going to talk too much about the statistic because it's not worth our time at this point in the course to go through this the, the mathematics of it. Epidemiologists in, in talking about the accuracy with which you predict something like a disease outcome talk about a, a statistic called area under the curve. An area under the curve varies between 0.5 or 50% and 1, a 100%. 50% is just chance, it's like flipping a coin is one way of thinking about it. A 100% is that you're perfectly accurate. And predicting somebody will have the disorder. In general, you don't expect perfect accuracy the general guidelines that epidemiologists use is that if this statistic is 75% or greater than it's a useful statistic for screening populations to make individual decisions or diagnosis, you would want it to be above 99%, a, a very high standard to me. So, if we look at current GWA standards and we look at the statistic, does it approach 75% or 99%? Well, as you might imagine, given that most heritability is missing, we're not going to approach those standards. This is a, a, a paper published by Chun Do where he compares the, the extent to which you could predict from a family history versus what you found from GWAS for this various disorders. And what you can see is that except for macular degeneration, none of these disorders are getting above that even 75% threshold. They're better than chance, so there is some predictive utility both from family history which is blue here and the, the, the snips, but it will be hard to argue that at this particular point in time, you're getting a lot of predictive utility out of the GWAS, at least at this level. So for not getting much, what's the reason to be optimistic that we'll ever get any utility from genetic prediction. Well I've, I think I'm inherently an, an optimistic person. So I want to give you three reasons why I think I, I'm optimistic and maybe why, hopefully, some of you will also be optimistic. Reason number 1 is that, and I've already alluded to this earlier in this course, is that even though overall and that area under the curve is giving you an overall predictive utility of what you're finding in GWAS, or just knowing someone's family history. Whether or not they have an affected relative, even though over all it might not be particularly useful. For some individuals and some cases, we know genetic prediction is very important. And we know more and more about these every day because of things like genome sequencing. So most of these revolve around rare variants. So we talked about this when we talked about schizophrenia. 30% of individuals who inherit a deletion on chromosome 22 something I called velocardiofacial syndrome before but more broadly called 22q11.2 deletion syndrome, 30% of those individuals we know would develop schizophrenia. So there is a case where we do have some predictive utility or individuals who inherit a deletion on chromosome 16. We know, it appears that almost all of those individuals will suffer an intellectual disability. The majority will have autism or autism like symptoms. We also know of genetic, rare genetic variants, variants that maybe only 1 out of 10,000 people in the population carry that affect risk of disease. Most of those are for neurological diseases at, at this particular point in time. Things like Alzheimer, dementia, where rare mutations in the amyloid precursor protein or the presenilin genes are associated with a vastly increased risk for suffering this neurological disorder. So even though overall we might not predict, in some cases we can begin to predict. Most of these revolve around rare variance. These are all things that don't happen, that often. And we might think, well if they don't happen that often, maybe we shouldn't care about them. That's why I put this last bullet point here. We all carry multiple rare genetic variants that when combined with a, a, another rare genetic variant at the same locus in a recessive condition will lead to a debilitating disorder, congenital disorder in a child. Roughly, the estimate is roughly 1 out of every 25 couples are in a situation where both members of that couple are carrying a rare recessive variant in that they then have a chance of producing a child that will carry or will express that rare disease. So, overall they might be rare, but they're still going to impact quite a few of us. That's the first reason I'm optimistic. The second reason I'm optimistic is that,. I think we're at the early stages of finding genetic variance, and that over time for sure we, the, and I tried to illustrate this in the case of schizophrenia, if you look at the first three major meta analysis of GWA studies of schizophrenia the first one found virtually nothing, the second I forget exactly how many 10, 20 variance, and the last one over a 100 variance. What will the next find? Maybe, 200, 300. So over time we're going to find more and more variance. And that should increase our predictive utility. How much will it increase our predictive u, utility? Again, a paper by nao, Naomi Wray, that geneticist from Australia I mentioned a little bit earlier in this module. Here, I'm going back to this is the, this, the blue bar is family history, whether or not you have a first degree relative with the disorder. The red bar is the predictive utility of what we know now from GWAS. And the green bar is what Naomi Wray estimates we could know if we could account for all that missing heritability. Maybe that's 50, 60 years down the road. I'm not sure. And again, the metric here is this area under the curve. 50% being chance. Anything above 50% has some predictive utility. But we'd like to get at least above 75%. So that we have screening potential ideally above 99%, but that's going to be extremely hard. That would mean that you could actually diagnose somebody from a genetic test. For these disorders it doesn't appear that you can be able to diagnose with great accuracy. But what Naomi Wray is showing is that over time maybe we will be able to get at a level of predictability for all these disorders. If we can account for that missing heritability. It's obviously going to take a lot of resources, but the predictive utility is at least in principle there. Now when faced with this data, one thing I think it's reasonable to think about. Is well G, you know, well, well people start using this type of information on genetic prediction to make reproductive choices that we might be a little bit uncomfortable about. That is maybe they decide to terminate a pregnancy because there is some genetic prediction for some trait that suggest that the, the, the fetus might carry an increased risk for that trait. Now people are going to have different opinions about that. Which is fine. But what I want to highlight here is really the gap between what epidemiologists feel is the standard you need to make, to make individual predictions at an accurate level. That 99% versus what Naomi Wray is reporting here as the best we might be able to achieve with genomic prediction. We'll never achieve perfect accuracy with genomic prediction, right? Because none of these disorders are perfectly heritable. The environment is important, and do you accept the environment is important, you can never fully accurately predict. And I think because of that, and I think this is important lesson for people to recognize. When thinking about genetic prediction especially as it might intersect or interact with reproductive decisions is there's always going to be a fair degree in, of inaccuracy. Because even if we could it fully account for the missing heritability, the heritability isn't one, and I'm going to illustrate that with predicting for IQ. So, a quantitative trait, which is something that people worry about. Will if, if we accounted for all the heritability of IQ, will people use that to predict the likely IQ of a fetus, and then based on that make reproductive decisions? Again I, it, it's not my role. I don't think to so much judge the moral the moral value or the moral legitimacy of doing that. What I'd like to highlight here is in my opinion, I think it's going to be very hard, even in the best case scenario, to make that prediction, because IQ is not perfectly heritable. If IQ is 60% heritable, which is probably the ballpark of what it is, then the error prediction from, if we knew all the genetic variants that accounted for that 60% would be plus or minus 19 IQ points. That's not a very accurate prediction. And even if it were as high as 80% heritable, there'd still be a fair degree of inaccuracy of predicting an individual's IQ just knowing his or her genotype. The last reason I mopped in this think about genetic prediction is that even though now we can't predict very well at the individual level. We can begin I think maybe within the next few years of identifying a subsets of the population that are extremely elevated risks for developing a disorder or at their very low risk for developing disorder, and to illustrate this I want to go back to the schizophrenia GWAS that we talked about in module 5. In this case, it's a kind of a complicated slide. Forgive me, so give me a second to explain what they did. Just recall that study. In that study, it was a massive study, there are about 150,000 people that were meta analyzed in that GWAS, and it was a case controlled study, and it maybe about 35 to 40,000 individuals that's schizophrenia. The rest were non, a people with non, who did, who did not have schizophrenia. And they found in their Manhattan plot 108 different regions of the genome, snips that were associated with risks for schizophrenia. Of course, no one of those snips is very predictive. All of them had a very small effect on your risk of schizophrenia. But what they could do is add up how many of the 108 risk variants a given individual had, and then ask if you had a high number of those 108 risk variants, what was your likelihood of developing schizophrenia versus if you had a low number of those risk variants. And that's what they're doing in this figure here. They actually, in, in, in effect what they did is they took those 108 variants, and they created a score for each individual in three different samples. There are three different, one from Denmark, Sweden and I forget what the third one is. But they're all very similar results. They created a score for each individual in the sample. Based on how many of those risk variants they had, and then they just divided up the sample into tenths, or decile. So these individuals down here, the lowest decile, had very few of those 108 risk variants. And basically what's plotted here, we could think of it this way, it's actually the odds ratio. But we can think of it as equivalent here, to their risk of schizophrenia. It's not to far off from what's going on here, for us to conceptually interpret what, what their results. The individuals at the lowest end here, had a very, very, low risk of developing schizophrenia. But look at the individuals who are in the top 10% of the sample. They carried a lot of the risk variance. They had a high risk of developing schizophrenia. It wasn't a 100% right, it was more like 15, 10, 15, 20% risk of developing schizophrenia. That's today, maybe the next, when they do the next large GWAS of a much larger sample, maybe they'll be able to refine this risk profile even more. The reason I think that's important now is that this group here, this highest 10%, actually represents a sub population that we might target prevention efforts at. I think there's a lot of things that we'd need to discuss in doing that, and in part one of the things we'd want to discuss, or we'd want medical professionals to dis, discuss, is the extent to which we stigmatize people by targeting them in this way. But nonetheless, I, I think we, we should and can have those conversations, and I think, ultimately, what we might see is that even though as we go along,. Maybe we won't be able to predict overwhelmingly accurately at the individual level, we'll be able to begin to identify sub groups of the population who are at high risk for developing schizophrenia, depression, diabetes, obesity, and we can target scarce resources for prevention on those individuals. So it's for those three reasons, already because of rare variance, we could begin to engage in genetic prediction for some individuals. Secondly, prediction is only going to get better as the GWAS findings as GWAS samples get larger. And third, even now, we can identify sub sub-populations where we might target our prevention efforts. I think even though genetic prediction is obviously a very challenging task, I think there's reason to be optimistic. And I've hope I convinced some of you as well. Thank you.