Integrating Protein Flexibility into Machine Learning Models for Virtual Hit Identification

• Grant number: 24/01956-5
• Support Opportunities: Regular Research Grants
• Start date: September 01, 2024
• End date: August 31, 2026
• Field of knowledge: Biological Sciences - Biophysics - Molecular Biophysics
• Agreement: CNPq
• Principal Investigator: Raghuvir Krishnaswamy Arni
• Grantee: Raghuvir Krishnaswamy Arni
• Host Institution: Instituto de Biociências, Letras e Ciências Exatas (IBILCE). Universidade Estadual Paulista (UNESP). Campus de São José do Rio Preto. São José do Rio Preto , SP, Brazil
• Associated research grant: 20/08615-8 - Protein exosites, cryptic sites and moonlighting: identification, functional mapping and effects of changes in structure, AP.TEM
• Associated scholarship(s): 24/13327-2 - Integrating Protein Flexibility into Machine Learning Models for Virtual Hit Identification, BP.JD

Abstract

In recent years, the expansion of chemical libraries beyond 10E8 compounds has opened new avenues for discovering hits against diverse targets. However, the sheer size of these libraries poses a challenge for traditional virtual screening (VS), particularly for academic researchers, given the substantial computational resources required for this undertaking. In response to the previous challenge, machine learning (ML) methods have emerged as a pivotal advancement, efficiently inferring virtual hits from docking results of a small fraction of the chemical library. Despite this progress, the current state of ML-driven VS approaches leaves room for enhancement, particularly in addressing factors that are key to boosting VS power. In this context, the current proposal aims to explore innovative strategies for integrating protein flexibility into existing ML-driven VS methods. Our approach involves constructing ML-classification models capable of predicting, from an ensemble of protein conformations, the most likely conformation to maximize docking scores for each compound. This novel methodology seeks to alleviate the computational burden associated with brute-force ensemble VS. To validate the efficacy of the ML models to be developed, extensive benchmarking will be conducted, followed by prospective studies focusing on allosteric sites of proteins of interest, a current focal point in our research group. The identified virtual hits will undergo rigorous experimental validation, contributing not only to the assessment of the proposed approaches but also holding the potential to unveil hits against pharmaceutical targets of significant interest. (AU)