Polygenic Risk Scores and SNP (a variation in a single-base pair in a DNA sequence)Heritability● Last five years○ Twin studies have mostly been done○ We know everything is heritable○ Linkage and candidate genes have shown there are not genes for anything○ GWAS has found a handful of tiny associations that may or may not meananything○ All this about trying to translate heritability into biology, causation● Polygenic Risk Scores: (It serves as the best prediction for the trait that can be made whentaking into account variation in multiple genetic variants.)○ Take the million or no SNPS, order them by association w/ outcome , and addthem together then the possible scores are 0,1 or 2, weight by effect size○ Correlated .1-.3 with behavior, .5 with height (retreat from biology)○ Twins are correlated .9 for height, .6 for IQ, .4 for educational attainment (itmeans that you can do genetically informed work in people without twins○ We could make an environmental PRS (big survey data, get a bunch of binaryindicators of educational attainment)● Dunedin Study○ 1,037 young people from NZ, followed throughout their lives○ PRS estimated in much larger samples, then applied to smaller sample like PRS● Belsky○ Educational Attainment r=.15○ Important to realize that this is a VERY small correlation○ But, EA also correlates with Parents (r=.13), learning to read etc earlier, higheraspirations, grades, more financially planful, richer spouses, more self-control,and not healthier○ Associated with Variance in breastfeeding, smoking during pregnancy, “Parentalsmacking”, income, television, and maternal education● Problems w/ PRS○ Associated with characteristics of your parents and your environment○ One way to get around is to study “mobility”: how much your achievementchanges from that of your parents● Sibling Difference Analysis○ Conducted within pairs of DZ Twins (randomized for PRS)○ Between pair correlation in DZ twins (r=.32) within pair correlation (r=.13)● Genome Wide Complex trait Analysis○ Final step away from biological action of genes○ A way to compute heritability without twins○ Remember how twin heritability works○ Big matrix of “unrelated” people○ Use SNP chips to establish how genetically similar they are○ Relation between the two is heritability○ No reference to action of individual genesThe Missing Heritability Problem- Published by Brendan Maher (Science journalist, notscientist)● Core problem in everything we’ve been studying this semester○ Height is 90% heritable○ But all the GWAS genes for height together add up to maybe 5% of the variance(then) and 20% (now)○ Worse in behavioral sciences then and now● Two Aspects for Maher○ Missing prediction problem - twins and parents are great predictors, genes not somuch○ Missing mechanism problem - how do the genes, when we find them, actuallywork to explain anything?● Where’s the Missing Heritability?○ Genetic structures not captured on SNP chips, Copy Number Variations, Raremutations, Gene-Gene interactions, Epigenetics○ New Aspect is GCTA○ The original MHP was about gap between heritability estimates and eitherprediction (PRS) or explanation (biology)○ GCTA allows us to compute heritability from SNPs, without twins○ SNP heritability is lower than twin heritability○ So a new version of MHP is between heritability coefficients from twins and SNPs● The Missing Environment Problem○ Remember that the same problem exists for the environment○ We know that “Nonshared Environment” accounts for a lot of variance○ We know that “Poverty” Accounts for a lot of variance● Hierarchy○ Linear prediction higher than biological explanation, because we are blindlycombining “effects” of DNA○ Why is machine learning higher than linear models, because we are combiningeven more blindly○ Why in SNP heritability higher than prediction, because we aren’t predicting, weare just modeling similarity○ Phenotypic heritability greater than SNP heritability, because relatives are REALdevelopment, not models of development● Big Question 1: How Much Better Can We Get?○ Biological explanation- very difficult, because of physics of carpets and inability todo experiments on humans○ Linear prediction models- to some extent, larger samples give us better models,but there’s a limit on how well linear models can perform as a model of realdevelopment○ Nonlinear prediction models- Statistical and computing advances, big samples,but still it’s a model○ GCTA- Better chips, whole genome sequencing● Big Question 2: Does Prediction Equal Determination?○ Let’s say we got to the point where we could predict IQ at r^2=.5 inEuro-Americans○ Does that mean that half the variance in IQ is determined at birth? Yes○ Prediction across groups doesn’t work very well● How This is Different from Eugenics 1.0○ Based on in vitro fertilization embryos○ Would allow parents to genotype and test embryos○ Already used spot rare diseases○ Or sexIntelligence● Genetics of Intelligence○ In many ways the core problem of any discussion of genetics and behavior○ Goes back to Galton○ The one phenotype that really matters theoretically○ Along with maybe schizophrenia● Genetics of Intelligence○ Behavior genetics and intelligence have something in common○ Both can easily lead to racist, regressive or even Nazi type outcomes○ This makes it tempting to dismiss them as completely false, to protectprogressive ideals○ But that’s too easy - We have learned that genetics of behavior does havemeaning, but it isn’t deterministic and the concept of intelligence has meaning,but it isn’t a fixed human capacity● Origins in Craniometry○ Belief that smarter people have larger heads/ brains than less intelligent○ Lots of roots in 19th Century racism○ There are small correlations between intelligence and head size - But there couldme a million reasons why● Alfred Binet○ Started out in craniometry○ Wanted to measure intelligence○ Identify children who need special education○ Created first modern IQ test○ Mental age○ Validation, development, subtests● H. H. Goddard○ Eugenicist○ Used Binet test to measure something called “innate intelligence”○ Coined the term moron○ Saw the proliferation of “high grade morons” as a major social issue● Lewis Terman○ Mass marketed the Binet scales○
View Full Document