Remo Rohs
Affiliation
University of Southern California, USA - Rohs Lab
Integrating deep learning and physics-based methods to study molecular interactions
Abstract
Research on molecular interactions increasingly relies on the ability to generate large amounts of data in biological experiments and on the exponential growth of computing power. The combination of data and computation forms the basis for the recent development of AI-based computational biology methods. My lab develops such tools based on molecular structure, with the goal of answering biological questions related to gene regulation, nucleic acid structure, protein-nucleic acid binding, and drug design. I will discuss DeepPBS, a method for predicting protein-DNA binding specificity from structural data, and DrugHIVE, an approach for designing drug-like molecules that are not available in current drug libraries. I will further discuss a hybrid pipeline that integrates AlphaFold predictions and molecular simulations with our DeepPBS method. This multiscale framework successfully captures conformational flexibility and the modulation of DNA structure through protein binding, providing new insights into transcription factor binding specificity and creating a roadmap for integrating deep learning and physics-based methods to study molecular mechanisms.