Discovering differential genome sequence activity with interpretable and efficient deep learning

Discovering sequence features that differentially direct cells to alternate fates is key to understanding both cellular development and the consequences of disease related mutations. We introduce Expected Pattern Effect and Differential Expected Pattern Effect, two black-box methods that can interpret genome regulatory sequences for cell type-specific or condition specific patterns. We show that these methods identify relevant transcription factor motifs and spacings that are predictive of cell state-specific chromatin accessibility. Finally, we integrate these methods into framework that is readily accessible to non-experts and available for download as a binary or installed via PyPI or bioconda at https://cgs.csail.mit.edu/deepaccess-package/. Author SummaryWithin the genome are the instructions to build all the cell types that make up the human body. However, understanding these instructions and how and when these instructions go wrong in cancer or genetically inherited disease is an open problem. Deep neural networks provide powerful models to learn the relationship between DNA sequence and functional consequence across many different cell types, such as whether a particular stretch of DNA is accessible and genes in that region can be expressed or is inaccessible and therefore genes are inactive. Despite these advances, a major setback in deep learning is that it is challenging to understand what patterns of DNA sequences a deep learning model has learned to associate with a particular genomic function, whether these patterns are significant, and how to determine whether these patterns are specific to a particular cell type or are general "housekeeping" patterns that function across many cell types. We introduce Expected Pattern Effect and Differential Expected Pattern Effect, two methods which allow us to evaluate the significance of particular patterns of DNA sequence features on models trained to predict function across multiple cell types, and apply this to problems of transcription factor binding and DNA accessibility across multiple cell types.