Pleiotropy Informed Genetic Risk Prediction via Logistic Multi-Task Learning

• Activity Number: 289
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Cong Li*+ and Can Yang and Hongyu Zhao
• Companies: Yale and Yale and Yale
• Keywords: Genetic risk prediction ; GWAS ; logistic regression ; multi-task learning ; pleiotropy

Abstract

Building risk prediction models for complex human diseases from genetics data is an important yet very challenging task of human genetics. The performance of genetic risk models is largely bottlenecked by the polygenic architecture of complex human diseases and the limited sample sizes of genome-wide association studies. A promising strategy that has emerged recently to improve genetic risk prediction is to leverage ``pleiotropy'', the shared genetic factors underlying related diseases. However, current methods in the literature are either not designed for handling disjoint sets of individuals for different phenotypes or not ideal for binary phenotypes. To account for these limitations, we propose a logistic multi-task learning model to leverage the pleiotropy between complex human diseases. In both simulations and real data, our method achieved similar prediction accuracy with bivariate ridge regression. However, compared with the bivariate ridge regression model above, our model is a logistic model and therefore deals binary phenotypes more naturally and provides interpretable estimated disease risk for each individual. Moreover, parameter tuning is much easier for our model tha