Seminar Series: Integrative analysis of multi-omics data improves genetic risk prediction and transcriptome-wide association analysis

Accurate prediction of disease risk based on genetic factors is an important goal in human genetics research and precision medicine. Better prediction models will lead to more effective disease prevention and early treatment strategies. Despite the identification of thousands of disease-associated genetic variants through Genome Wide Association Studies (GWAS) in the past decade, genetic risk prediction remains a challenging problem, mainly due to the enormous amount of weak features and lack of individual-level training data. In my presentation, I will first introduce a statistical framework to predict disease risk using GWAS summary statistics as training data and integrates diverse types of external information. Through comprehensive simulations and real data analyses, we demonstrate that our approach can substantially increase the accuracy of risk prediction and population stratification. I will then describe a multi-task learning approach to jointly predict gene expression levels in 44 human tissues and its application under the transcriptome-wide association analysis framework (TWAS). Compared with prediction models of 15,000 genes trained separately in each tissue, our approach achieved an average 39% improvement in prediction accuracy and generated an average 120% more effective prediction models. To demonstrate the merit of joint modeling in downstream TWAS, we applied our framework to 50 complex traits (Ntotal=4.5 million) and were able to identify considerably more