Analysis and Integration of Multiple Omics Traits by Joint Variable Selection via the Elastic Net

Genome wide association studies have provided valuable information about human variation and complex trait regulation, but remain difficult to interpret due to the genetic context that the majority of the human genome lacks. Previous efforts to explain complex trait regulation have focused on leveraging single-omics traits such as the transcriptome to predict biologically relevant features based on genomic data. This provides valuable insight into genetic therapeutic targets. In this study I analyze the results of univariate protein and RNA model building via the elastic net and discuss one method of integrating these datasets by joint variable selection via the elastic net. I utilize data from the Trans-Omics for Precision Medicine (TOPMed) and Multi-Ethnic Study of Atherosclerosis (MESA) consortiums which provides measured genotype, protein expression and RNA expression for 388 individuals of self-identified European descent. I demonstrate that when univariate modeling, only 30% of models built meet a threshold of R2>0.1 in both protein models and RNA models. This is reflected in joint models which only demonstrate a weak correlation of model R2 (rho=0.14) between protein and RNA models. In out of sample testing there is no significant difference in the performance of univariate response protein models and protein models built jointly with RNA models (p>0.7 for all out of sample tests). This study demonstrates that many omics traits can explain more human variation than single omics traits and explores one method of integrating this data that could provide valuable insight given more development.