A0731
Title: Leveraging genetic correlation for multi-trait polygenic scores construction via L1 penalized regression
Authors: Osvaldo Espin-Garcia - University of Western Ontario (Canada) [presenting]
Abstract: Polygenic risk scores (PRS) quantify the genetic contribution of an individual's genotype to a trait, e.g. disease or phenotype. PRS can be used to group subjects into different risk strata and can thus be treated as predictors in clinical and epidemiological studies. However, PRS methods typically focus on a single trait at a time, ignoring the potential simultaneous influence of genes on multiple traits. To address this limitation, a recently published model that uses an L1 penalty is extended by incorporating a genetic correlation matrix among traits into the cost function of the penalized regression framework. The main objective is to improve the predictive ability of multi-trait PRS models when multiple traits are of equal interest. The proposed method is evaluated against alternatives via comprehensive numerical studies with a focus on marginal and joint performance metrics. The penalized approach is applied to two studies: one analyzing smoking and drinking behaviors in a cohort of head and neck participants and another one examining cardiovascular traits from a birth cohort from Finland. Lastly, a memory-efficient re-implementation of the penalized regression framework will be discussed.
