intDesc-AbMut: Describing and understanding how antibody mutations impact their environmental interactions

Abstract

Previously, we proposed a double-point mutation (DPM) strategy involving the simultaneous substitution of two amino acids to optimize antibodies. By selecting mutants based on the criterion that favorable interactions between mutated residues and their local environments are preserved or enhanced, improvements in antibody affinity were achieved. Nonetheless, manual extraction of these interactions proved to be time-consuming and labor-intensive. Thus, to streamline this process, we developed intDesc-AbMut, an automated software tool designed to extract interactions between designated mutant residues and their surroundings from three-dimensional antigen-antibody complex structures. intDesc-AbMut identifies and classifies 36 distinct types of interactions, enabling visualization of changes before and after mutation. Additionally, the effects of mutations can be represented by interaction descriptors generated from the extracted interactions. As a practical application, we constructed a machine learning model using these descriptors combined with side-chain rotamer frequencies to determine whether the environment around mutated residues in antigen-antibody complexes is crystal-like, achieving a Matthews correlation coefficient of 0.505 and an accuracy of 0.855. The intDesc-AbMut software is available at: https://github.com/riken-yokohama-AI-drug/intDesc/tree/intDesc-AbMut.