Abstract
Structure-based antibody engineering requires systematic and reproducible characterization of residue-level interactions within antigen-antibody complexes. In our previous double-point mutation strategy, mutant selection was guided by preservation or enhancement of favorable local interactions; however, identifying and comparing these interactions depended on labor-intensive manual structural inspection. Here, we present intDesc-AbMut, a residue-centric computational tool that automates the extraction, classification, visualization, and descriptor generation of interactions involving designated residues in antigen-antibody complexes. The software explicitly defines 36 interaction types, encompassing conventional hydrogen bonds and van der Waals contacts as well as weak hydrogen bonds (e.g., CH···O and CH···π), sulfur-related contacts, bond dipole interactions, and orthogonal multipolar interactions. This fine-grained interaction taxonomy enables quantitative representation of local packing, weak hydrogen bonding, and directional electrostatic features surrounding mutated residues. As a proof of concept, we incorporate these interaction descriptors into a machine learning framework to evaluate whether they encode structural characteristics consistent with experimentally observed side-chain conformations. The results demonstrate that the descriptors capture meaningful structural signals underlying crystal-structure-like conformations. This analysis assesses descriptor informativeness rather than predicting binding affinity. In summary, intDesc-AbMut provides a systematic and extendable framework for mutation-focused interaction analysis in antibody structural studies and related protein design applications.