Predicting allostery in proteins

Abstract

Allosteric site prediction has emerged as a pivotal approach in drug discovery, offering enhanced selectivity and reduced side effects compared to traditional orthosteric targeting. This report explores multiple computational strategies for identifying allosteric sites, with a focus on G protein-coupled receptors (GPCRs). Among the surveyed methods, machine learning models like PASSerRank demonstrate exceptional performance by employing ensemble learning, automated optimization, and ranking algorithms to prioritize potential allosteric sites with high accuracy. The solution implemented integrates structural and sequence filtering, feature ex- traction, and advanced predictive modeling, achieving robust results by leveraging techniques such as PDB data curation, fpocket-based pocket detection, and cheminformatics-driven feature analysis. Our workflow highlights the importance of integrating machine learning with domain- specific constraints to refine predictions and pave the way for the discovery of novel therapeutic targets.