With an intention to develop a persuasive
With an intention to develop a persuasive model for the screening of potent molecules as Cyclin dependent kinase inhibitors, 2D QSAR studies using CANVAS and 3D QSAR studies were carried out on 230 CDK4 inhibitor molecules with pyrido [2,3-d]pyrimidine moiety. By using CANVAS, several statistical analysis was carried out using both fingerprints and molecular descriptors. It is found that kernel based partial least square analysis fitting with fingerprints using 15th split of the learning set into a test set and training set created a worthy model, among which the one with molprint2D fingerprints generated a noble model with a score value of 0.8322. External validation of the molecules were also showed a high degree of correlation with observed biological activity. Predictive power of the model is tested with test set 2 and also with Decoy set of 75 molecules. As a result we got above 0.8 correlations with the kpls_molprint_2D model. But more effective predictions are obtained by using the consensus model generated by combining kernel based partial least square models of 2D molprint, linear and radial fingerprints (correlation is 0.92). An ptio based 3D QSAR model was generated from the same sets of molecules. Atomic effects of each molecule for the activity were also realized. Number of PLS factors are restricted to 6. A highly predictive model with R 2 = 0.8372 and Q2 = 0.7381 was obtained. The usefulness of the model was understood by its cross-validated R2 value which is found to be 0.5943 and Y scrambling (0.4470). From the study, it is noticed that presence of hydrophobic or nonpolar region along with electron withdrawing features in specified areas stimulate inhibitory activity of molecules. Docking experiments of these molecules in the extra precision mode with ligand flexibility followed by the binding energy determinations shed light on the leading non-covalent interactions causative of the inhibitory activity of CDK4 inhibitors. From the study, it is found that hydrogen bonding interactions with hinge region residue Val96, and residues Asp99, Lys35 and Thr102, salt bridge formation with Asp99 and л-л stacking interaction with His95 and Phe93 are common in active molecules. This л-л stacking interactions are found in the hydrophobic or non-polar region of the ligand molecules. Presence of carbonyl group in the hydrophobic area has a tendency to reduce the inhibitory activity. In molecule 187, salt bridge formation is observed in the electron withdrawing area. This area is also specified by the positive ionic features. The stability of the protein-ligand complex can be explained by calculating the decrease in binding energy after docking. The major factors that were influencing the stability of protein-ligand complex are the van der Waals interactions, lipophilic interactions and coulombic interactions. Thus, the consensus prediction of molecules using 2D QSAR models developed though kernel-based partial least square method using fingerprints besides the atom based 3DQSAR model is highly effective for screening a set of molecules with various scaffolds.