BioML Seminar 5.2 - Decoding the Grammar of Regulatory DNA

​Another week, another seminar from the BioML group in Machine Learning at Berkeley, sponsored by Amplify Partners. This week, we're excited to host Chang Yun, a Ph.D. Candidate at Stanford University.

​Abstract: Over the last two decades, the Encyclopedia of DNA Elements (“ENCODE”) Project has performed diverse genomics experiments to map millions of functionally active regions in the human and mouse genomes that control when, where, and how much nearby genes are expressed. However, the experimental maps do not fully explain how this regulatory activity is encoded in the DNA sequence itself: which individual DNA bases are important, how combinations and arrangements cause a region to behave differently across cell types, or how a genetic variant might alter activity. To help answer these questions, we developed ENCODE GRAMMAR (Genomic Regulatory Atlas of sequence Models, Motifs, Annotations and Rules): one of the largest collections regulatory DNA sequence models to date (also colloquially called “sequence-to-function”, or “seq2func” models), across 3,865 experiment datasets that span multiple layers of gene regulation, with accompanying model interpretation frameworks that identify the sequence features and rules that drive model prediction. With ENCODE GRAMMAR, we show how model predictions and interpretations from multiple models and modalities can be integrated to decode the sequence basis of regulatory activity, at scale.

​Speaker Bio: Chang is a Ph.D. Candidate at Stanford University, advised by Anshul Kundaje and co-advised by Brian Hie, working on the molecular mechanisms of gene regulation and the interface of protein-DNA interactions. Prior to his Ph.D., he received his M.Phil. in Biotechnology from the University of Cambridge, and his B.S. in Chemical Engineering from Columbia University.