“…Software packages to probe the structural details of protein-ligand interactions do not allow the study of many complexes at once and are limited to only certain types of ligands (Adasme et al, 2021;Laskowski et al, 2018;Laskowski & Swindells, 2011;Schöning-Stierand et al, 2020;Young et al, 2022). And while protein structural search methods such as TM-align (Zhang & Skolnick, 2005), Dali (Holm, 2020), andFold-Seek (van Kempen et al, 2024) have enabled the identification of proteins with local structurally homologous regions, and programs like SSDraw (Chen & Porter, 2023) and ConSurf (Ashkenazy et al, 2016) can align sequences to show how amino acids (AAs) and secondary structures are conserved in a set of homologous proteins, these methods are not designed to highlight or organize conserved binding interactions across diverse proteins: for example, small molecule ligands that bind multiple proteins via different binding pocket geometries or antigens that bind a library of engineered antibodies at non-overlapping epitopes. In such cases, translating trends at the 2D sequence level to the 3D space in related proteins toward identifying molecular patterns that drive or tune biochemical interactions presents a challenge for which no generalizable software is available, even when high-confidence interactions can be predicted at a large scale (Bryant et al, 2022;Evans et al, 2022;Hwang et al, 2017;Petrey et al, 2023;Trudeau et al, 2023;Zhang et al, 2012).…”