Integrative approaches for predicting protein interactions and applications

University Of Missouri-Columbia, Columbia MO

Investigators

Linked publications & trials

Abstract

The ability to accurately predict protein interactions is critical for the mechanistic investigation of many biological processes and for rational therapeutic development. Over the past decades, methods ranging from physicochemical-based approaches to template-based and machine learning-based approaches have significantly advanced protein complex structure prediction. With the power of recent tools like AlphaFold2 and AlphaFold3, enormous progress has been achieved. Yet, there are several unmet, important challenges that remain unsolved beyond the existing computational methods, including AlphaFold. We will take advantage of the flexibility of the five-year R35 funding mechanism to address five different challenges under the umbrella of a common theme: developing novel computational methods to predict protein interactions and their complex structures by integrating principles of physics, bioinformatic approaches, and machine learning technologies. I will also continue to collaborate closely with experimental colleagues at our university, validate our computational methods and predictions, and explore potential therapeutic applications for the treatment of diabetic wounds. Goal #1: Predicting disordered protein-protein interactions. Goal #2: De novo design of peptides targeting the protein-protein interaction (PPI) interface. Goal #3: Development of novel peptide inducers of NRF2 that target the Keap1-NRF2 interface with potential applications for treating diabetic wounds. Goal #4: Development of a Graph Neural Network (GNN)-based deep learning algorithm for antibody-antigen interface True/False classification and multiple point mutation assessment. Goal #5: Development of GNN-based polarized protein-specific charges for protein-ligand docking and simulations. During the past four years since the funding of this R35 in May 2020, my lab has published 20 peer-reviewed papers supported by this grant on predicting protein interactions and lead discovery, including papers in high- impact journals such as PNAS and Nature Communications. I have additional five research manuscripts submitted to peer-reviewed journals (currently under review). We have also filed three pharmaceutical patent applications and obtained one software copyright, along with three webservers. Our software has been freely distributed to academic users from 22 countries across every continent. My team consistently ranks among the top in prestigious and highly competitive international competitions for blind protein-protein and protein-ligand structure predictions. All source codes developed from this project will be freely accessible to academic users and all webservers are freely accessible to everyone.