Machine Learning and Optimization Algorithms for Intra- and Intermolecular Interaction Prediction

Abstract

Computational prediction of intra- and intermolecular interactions, specifically intra- protein residue-residue interactions and the interaction sites in between proteins and other macromolecules, are critical for understanding numerous biological processes. The existing methods fall short in estimating the quality of intra-protein interactions. Moreover, the methods for predicting intermolecular interactions fail to harness some of the latest technological advancements such as advances in pretrained protein and RNA language models and struggle to effectively integrate predicted structural information, thus limiting their predictive modeling accuracy. Hence, my objectives include (1) the development of computational methods for protein structure modeling through the estimation of intra-protein interactions, (2) the development of computational methods for predicting protein- protein interaction sites leveraging the latest deep learning architectures and predicted structural information, and (3) extending the scope beyond protein-protein interactions to develop novel computational methods to predict protein-nucleic acid interactions informed by protein and RNA language models. The major benefits of achieving these objectives for the broader scientific community are the following: (1) intra-protein interaction estimation methods have the potential to enhance the accuracy of protein structure modeling, and (2) the methods for predicting protein-protein and protein-nucleic acid interaction will deepen our understanding of biomolecular interactions in cell, even when experimentally determined molecular structures are not available.