Predicting gene expression of yeast with deep learning-based sequence model vs linear regression model

Abstract

Predicting functional effects of enhancers in the gene transcription process has been an active area of research to understand complex gene regulatory mechanisms and interpret non-coding variants. Quantitative modeling of sequence-to-expression of organisms has wide applications ranging from improvement in bioenergy products to better understanding of complex diseases such as cancer. To demonstrate a sequence-to-expression modeling of a yeast called Issatchenkia Orientalis, a data-driven convolutional neural network is trained, validated, and interpreted, and a linear regression model is constructed to provide performance and result comparison. Both models are being evaluated on real data of gene expressions of Issatchenkia Orientalis and provide unique insights into underlying gene transcriptomic mechanisms in Issatchenkia Orientalis