Low-Data RecA Inhibitor Prioritization with Docking-Aware Potency and RAD51 Selectivity Models

Antibiotic resistance is a longstanding issue in biology, and resistance to critical bactericidal antibiotics such as quinolones has increased significantly due to bacterial mechanisms that repair DNA damage and accelerate mutation. RecA is a bacterial recombinase that coordinates homologous recombination and SOS repair, making it an adjuvant target for weakening repair and enhancing DNA-damaging antibiotics. The central difficulty of targeting RecA is the structural similarity with human RAD51, limiting the clinical applicability of current inhibitors. Here, we present a pipeline for identifying potent RecA inhibitors without sacrificing RAD51 selectivity. Multi-conformation docking supplies physics-informed features on which RecA and RAD51 potency regressors are trained. Then, candidate selectionidentifies potent RecA inhibitors from a curated source of Rad51 negatives, resulting in RecA inhibitors inactive against Rad51. Future work seeks to utilize a teacher–student surrogate distilled from multi-conformational docking results to expand the screening scope, and initial biochemical validation is ongoing.