An Integrated Molecular Modeling with MPNN, MD Simulations, MM/PBSA, ADMET, and SHAP Analyses to Identify GyrB/PqsR Inhibitors Against  Pseudomonas aeruginosa

Abstract Antibiotic resistance in Pseudomonas aeruginosa underscores the urgent need for new therapeutics targeting multiple bacterial pathways. In this study, a graph neural network‐driven framework is presented, integrating Message Passing Neural Networks (MPNNs) with atom‐level SHAP interpretability to predict antibacterial activity, complemented by clustering, molecular docking, and molecular dynamics simulations. From 96 structurally diverse candidates, two molecules‐ LIG61 (4‐[5‐(6‐acetyl‐5‐hydroxy‐4‐methyl‐2,8‐dioxo‐9,10‐dihydropyrano[2,3‐h]chromen‐10‐yl)furan‐2‐yl]benzoic acid) and LIG87 ([2‐[(2‐methylphenyl)methylidene]‐3‐oxo‐1‐benzofuran‐6‐yl] N , N ‐diphenylcarbamate)‐demonstrates inhibitory potential against DNA gyrase subunit B (GyrB) and the quorum‐sensing regulator PqsR. Docking reveals that both ligands establish key polar and hydrophobic interactions with experimentally validated residues, consistent with reported co‐crystal data. MD simulations confirm stable interactions of LIG61 and LIG87 with their targets, and subsequent MM/PBSA analysis yielded binding affinities of −9.94 and −16.89 kcal mol −1 for LIG61, and −17.84 and −29.23 kcal mol −1 for LIG87, toward GyrB and PqsR, respectively. Predicted acute toxicity indicates LD 50 values of 400 mg kg −1 for LIG61 and 500 mg kg −1 for LIG87, demonstrating a favorable toxicity profile. SHAP‐based atom‐level interpretation of the MPNN predictions further identified key hydroxyl, lactone, and aromatic moieties underpinning bioactivity. Together, these findings highlight computational framework capable of identifying dual inhibitors with atomistic insights, with the understanding that in vitro and in vivo studies are required for validation.