Deep Learning for Non-coding Variant Interpretation

Non-coding genetic variants are important drivers of trait variation in humans and other eukaryotic species. As a result, functionally characterizing non-coding variants is a major goal of modern genetics, with important implications for our understanding of the molecular basis of disease and evolution. Deep learning models are a powerful approach for predicting regulatory activity from genomic sequence and interpreting the cis-regulatory code. In this session, four leading scientists will present their work on leveraging such models to interpret the effects of regulatory variation. These talks will highlight recent advances in deep learning for variant effect interpretation, limitations in existing models, and strategies for improvement. Particular advancements that will be discussed will include improvements in model architecture and training, novel interpretation and variant scoring techniques, and emerging strategies to scale model training across populations and tissue types.

Learning Objectives

 *   Highlight challenges of deep learning models for non-coding variant interpretation.
 *   Design deep learning architectures trained on functional human genetic variation.
 *   Construct deep learning models capturing the syntax of cell type-specific gene expression.