Machine Learning Approaches to Studying Chromatin

Jeremy Bigness

My work focuses on the spatiotemporal regulation of gene expression across different cell types. Since the linear sequence of DNA remains the same in almost all somatic cells, tissue specific gene expression must arise due to factors extrinsic to sequence, particularly epigenetic factors. For multiple cell types, we are studying how both local epigenetic signals as well as the global three-dimensional architecture of the genome jointly regulate gene expression – with an aim toward not only prediction but also model interpretability and mechanistic insight. To accomplish these goals, we are using a class of machine learning models called neural networks to integrate multimodal data sets. If successful, our model will not only account for feature interactions on a local scale but also distal interactions due to the spatial organization of the genome.