SWFB and GA Strategies for Variable Selection in QSAR Studies for the Validation of Thiazolidine- 2,4-Dione Derivatives as Promising Antitumor Candidates

Abstract:

Objective: Thiazolidine-2,4-dione (TZD) are the well known anti-diabetic scaffold. Very recently, several TZD based anti-cancer agents have came into limelight for treating mutant cancer forms. In order to establish and understand the relationship of biological activity with that of physiochemical parameters associated with the structure, twodimensional (2D-QSAR), group-based (G-QSAR), and three-dimensional (3D-QSAR) were performed which may be useful for (medicinal) chemists in selecting the most suitable substituent for the development of more potent, effective and selective TZD based anticancer agents in future. Methods: A series of TZD derivatives were subjected to 2D-QSAR, G-QSAR, and 3D-QSAR studies. The following studies were performed using partial least square regression, multiple regressions and k-nearest neighbor methodology coupled with various feature selection methods, viz. stepwise forward backward (SWFB) and genetic algorithm (GA) to derive QSAR models which were further validated for statistical significance and predictive capability by internal and external validation. Results: The results were expressed for both SWFB and GA consecutively. The statistically significant best 2D QSAR model has r² = 0.90, 0.89 and q² = 0.86, 0.84 with pred_r² = 0.87, 0.82 for PLSR with whereas r² = 0.97, 0.91 and q2 = 0.95, 0.86 with pred_r2 = 0.86, 0.77 were predicted for MLR. G-QSAR model has r² = 0.92, 0.81 and q2 = 0.90, 0.76 with pred_r² = 0.77, 0.77 for PLSR whereas r2 = 0.92, 0.88 and q² = 0.87, 0.71 with pred_r² = 0.73, 0.87 were predicted for MLR. The 3D-QSAR studies were performed by using of kNN-MFA approach; a leave-one-out cross-validated correlation coefficient q2 = 0.85, 0.84 and pred_ r² = 0.94, 0.77 were obtained. Contour maps using this approach showed that steric, electrostatic, and hydrophobic effects dominantly determine binding affinities. The docking study revealed the binding orientations of these inhibitors at active site amino acid residues (ARG281 and ARG 852) of PI3Kα enzyme (PDB ID: 3ZIM). Conclusion: The present research represents an effort to recognize the necessary structural requirements of TZD derivatives to be potential anticancer agents.

Key words: G-QSAR, kNN-MFA, Thiazolidine-2,4-dione, 2D/3D QSAR, VLife MDS.