205W Poster - Population Genetics
Wednesday June 08, 9:15 PM - 10:00 PM

Authors:
Dandan Peng; Obadiah Mulder

Affiliation: University of Southern California, Los Angeles, CA

Keywords:
Complex traits

A polygenic score (PGS) is a weighted sum of an individual’s genotypes used to predict the individual’s value for a complex trait. Estimating the evolutionary history of a population’s mean value of one or more polygenic scores may rule out or support hypotheses about trait evolution in the population. One of the key components in estimating a PGS history is to estimate the trait-associated alleles’ frequencies. In previous work, Edge & Coop (2019) proposed three methods to estimate the historical time course of a population-mean PGS using estimated local coalescent trees encoded by an estimated tree sequence or ancestral recombination graph (ARG). Among Edge & Coop’s three estimators, one performs well under neutrality but is biased under natural selection (proportion-of-lineages). The other two are less biased by selection but very noisy (waiting-time and lineages-remaining). Here we report the performance of Edge & Coop’s methods using state-of-the-art ARG estimation methods (Relate and tsinfer), and we explore approaches to improving the estimators, including smoothing approaches for the two noisy methods and an expectation-maximization (EM) approach to correcting bias observed under selection for the other method. The improved methods will contribute to more accurate estimates of PGS histories and allow the testing of more specific hypotheses about trait evolution.