Machine learning and generative AI in the rational design of DNA gyrase-targeted antibacterials

DNA gyrase, a critical bacterial enzyme, was targeted using an AI-driven approach to accelerate antibacterial drug discovery. Machine learning (ML) models, including Gradient Boosting Regressor (GBR) and XGBoost, were optimized on pIC₅₀ data, with GBR achieving superior generalization (train R² = 0.84, test R² = 0.76). A Graph Convolutional Network Variational Autoencoder (GCN-VAE) generated diverse molecular scaffolds, validated by Tanimoto similarity. Out of 100 AI-generated drug-like molecules, 11 were identified as structurally unique, with one (denoted as DR7) exhibiting a predicted pIC₅₀ > 7, indicating potent inhibitory activity. Docking studies of DR7 with DNA gyrase identified a lead molecule with a binding energy of −7.44 kcal/mol and inhibition constant (K_i) of 3.52 μM. Key protein-ligand interactions, involving TYR86 and ASP87, were highlighted alongside electronic characterization of HOMO and LUMO distributions, elucidating binding potential. This integrative framework of ML, generative AI, and molecular docking offers a transformative path for developing DNA gyrase inhibitors and advancing antibacterial therapeutics.