Integrating traditional QSAR and read-across-based regression models for predicting potential anti-leishmanial azole compounds.

Abstract

Leishmaniasis, a neglected tropical disease caused by various Leishmania species, poses a significant global health challenge, especially in resource-limited regions. Visceral Leishmaniasis (VL) stands out among its severe manifestations, and current drug therapies have limitations, necessitating the exploration of new, cost-effective treatments. This study utilized a comprehensive computational workflow, integrating traditional 2D-QSAR, q-RASAR, and molecular docking to identify novel anti-leishmanial compounds, with a focus on Glycyl-tRNA Synthetase (LdGlyRS) as a promising drug target. A feature selection process combining Genetic Function Approximation (GFA)-Lasso with Multiple Linear Regression (MLR) was used to characterize 99 azole compounds across ten structural classes. The baseline MLR model (MOD1), containing seven simple and interpretable 2D features, exhibited robust predictive capabilities, achieving an R2train value of 0.82 and an R2test value of 0.87. To further enhance prediction accuracy, three qualified single models (two MLR and one q-RASAR) were used to construct three consensus models (CMs), with CM2 (MAEtest = 0.127) demonstrating significantly higher prediction accuracy for test compounds than the MOD1. Subsequently, Support Vector Regression (SVR) and Boosting yielded 0.88 (R2train), 0.86 (R2test), 0.92 (R2train), and 0.82 (R2test), respectively. Molecular docking highlighted interactions of potent azoles within the QSAR dataset with critical residues in the LdGlyRS active site (Arg226 and Glu350), emphasizing their inhibitory potential. Furthermore, the pIC50 values of an accurate external set of 2000 azole compounds from the ZINC20 database were simultaneously predicted by CM2 + SVR + Boosting models and docked against the LdGlyRS, which identified Bazedoxifene, Talmetacin, Pyrvinium, Enzastaurin as leading FDA candidates, whereas three novel compounds with the database code ZINC000001153734, ZINC000011934652, and ZINC000009942262 displayed stable docked interactions and favourable ADMET assessments. Subsequently, Molecular Dynamics (MD) simulations for 100ns were conducted to validate the findings further, offering enhanced insights into the stability and dynamic behaviour of the ligand-protein complexes. The integrated approach of this study underscores the efficacy of 2D-QSAR modelling. It identifies LdGlyRS as a promising leishmaniasis target, offering a robust strategy for discovering and optimizing anti-leishmanial compounds to address the critical need for improved treatments.