Hybrid APSO-Bayesian optimized triple-kernel SVR models for QSAR and rational design of 1H-pyrrolo[2,3-c]pyridine-based LSD1 Inhibitors.

Lysine-specific demethylase 1 (LSD1) is a critical therapeutic target for acute myeloid leukemia. This study developed quantitative structure-activity relationship models to predict the inhibitory activity of 1H-pyrrolo[2,3-c]pyridine-based compounds. Five models were established using heuristic method, XGBoost, and support vector regression (SVR). A triple-kernel SVR model, integrating Radial Basis Function, Polynomial, and Linear kernels, was constructed to capture linear and non-linear patterns, optimized via a hybrid Adaptive Particle Swarm Optimization and Bayesian strategy. Rigorous validation metrics demonstrated that this model achieved highly competitive generalization ([Formula: see text], [Formula: see text]). Interpretation of descriptors highlighted that the electronegativity of the nitrogen atom and hydrogen bond accepting capabilities are critical determinants for LSD1 inhibitory activity. Insights from this model guided the rational design of novel derivatives, whose favorable binding interactions were corroborated by molecular docking. To further explore these findings, 100 ns molecular dynamics simulations were performed on the lead candidate, Compound 46c. The complex exhibited high structural stability. MM/PBSA calculations revealed robust spontaneous binding affinity ([Formula: see text] kcal/mol), primarily driven by van der Waals interactions. Notably, trajectory analysis identified a persistent hydrogen bond with residue LEU183 as a critical stabilizer. Future work will focus on the synthesis and in vitro anticancer evaluation of the designed candidates, particularly Compound 46c, to evaluate their therapeutic potential.