At @herasight, we wanted to compare our genetic predictors (PGS) to those from @nucleusgenomics. However, in many cases, we couldn’t reconcile plausible performance of their PGSs with customer risk reports we saw — this may have misled customers about their disease risks. Nine of their PGSs appeared to be open-source models from PGS catalog. Many (see table) relied on small numbers of variants despite being for polygenic diseases. State of the art PGS typically use thousands or millions of variants to maximize predictive ability.

The table gives our liability scale R^2 (a measure of PGS prediction performance) back-engineered from customer reports, along with the number of variants used in the PGS, and the R^2 we achieved in our independent validation in UKBB.

I’ve been working with an IVF startup, @herasight, that has already screened hundreds of embryos. Today we come out of stealth with a paper showing that our predictors for 17 diseases — validated within-family — beat the competition, with improved performance in non-Europeans🧵 Check out our website where you can play with our widget and download our white paper: herasight.com

New educational attainment (EA) GWAS out today: nature.com/articles/s4158…. We expand the sample size from ~1 million to ~3 million, making it one of the largest GWAS to date. We identify more loci affecting EA and increase our ability to predict EA from genetic data, and more...🧵 First, some background: educational attainment (EA) measures the number of years of education an individual completes, information that is routinely collected by researchers in many fields, and is strongly related to many socioeconomic and health outcomes.

New from Richard Border @andywdahl @flint Noah Zaitlen myself & others. Estimated genetic correlations between traits may be inflated by cross-trait assortative mating (xAM): . So are apparent genetic relationships between traits a statistical artefact? 🧵biorxiv.org/content/10.110… The genetic correlation between two traits is usually interpreted as a measure of the degree to which the genetic effects on one trait are correlated with the genetic effects on the other, i.e. shared underlying biology. However, this interpretation is not valid when there is xAM