A0407
Title: Predictive modeling of transcription-wide association studies via statistical learning methods
Authors: Min Chen - University of Texas at Dallas (United States) [presenting]
Abstract: Traditional genomic and transcriptomic predictive models usually require strong assumptions on model structures and data distributions. In contrast, statistical learning models, developed mainly for the purpose of prediction, are less restrictive because they are able to learn with little model assumptions. This is very powerful because it allows for detection without specifying explicitly, e.g., whether the phenotype has additive/multiplicative, dominant/recessive, or epistatic effects. In addition, they can capture nonlinear structures defined by complex genomic and epigenomic regulatory networks. A statistical learning model is proposed to incorporate known regulatory relationships, pathways and epigenomic networks. The model is applied to GTEx and Geuvadis data to improve the prediction of risk and mRNA expression associated with SNPs.