An AI-Driven Platform for Deconstructing and Engineering Biomolecular Recognition

Abstract

Understanding molecular interactions at the atomic level remains a central challenge across biology, medicine, and engineering. We introduce Perturbation Scanning (PS), an interrogative AI framework that actively deconstructs molecular interfaces. PS integrates graph-based representations of structures derived from molecular dynamics trajectories or Protein Data Bank files to systematically probe the electrostatic, hydrophobic, and steric contributions of each residue. To translate these insights into actionable design, we introduce the Intelligent Interface Optimization Scanner (IIOS), a standalone tool that generates energy-scored mutation proposals from interface maps. Together, PS and IIOS provide an integrated platform for dissecting and rationally engineering molecular interactions by resolving force-specific and stage-dependent contributions that are not directly accessible with existing computational approaches. Unlike traditional alanine scanning or free-energy methods such as MMPBSA--which provide only static or ensemble-averaged measures--PS delivers stage-resolved, force-wise decomposition of binding interfaces via a suite of targeted physicochemical perturbations applied to its AI model, directly quantifying each residues mechanistic role.