Deep learning decodes the principles of differential gene expression.

Abstract

Identifying the molecular mechanisms that control differential gene expression (DE) is a major goal of basic and disease biology. We develop a systems biology model to predict DE, and mine the biological basis of the factors that influence predicted gene expression, in order to understand how it may be generated. This model, called <i>DEcode</i>, utilizes deep learning to predict DE based on genome-wide binding sites on RNAs and promoters. Ranking predictive factors from the DEcode indicates that clinically relevant expression changes between thousands of individuals can be predicted mainly through the joint action of post-transcriptional RNA-binding factors. We also show the broad potential applications of DEcode to generate biological insights, by predicting DE between tissues, differential transcript-usage, and drivers of aging throughout the human lifespan, of gene coexpression relationships on a genome-wide scale, and of frequently DE genes across diverse conditions. Researchers can freely utilize DEcode to identify influential molecular mechanisms for any human expression data - www.differentialexpression.org.