Dysregulation of P70 ribosomal S6 kinase (P70S6K) has been observed in many cancers; therefore, the design of new molecules targeting p70S6K of paramount importance in cancer therapy. The current study employed a group-based quantitative structure-activity relationship (GQSAR) to develop global QSAR models capable of predicting the bioactivity of P70S6K inhibitors. A wide variety of chemical structures and biological activities (half maximal inhibitory concentration) of P70S6K inhibitors were collected from the binding database website. Compounds were classified into various chemical groups and then fragmented into R1, R2, and R3 fragments based on certain pharmacophoric features required for ligand-target biointeractions. Different two-dimensional fragment-based descriptors were calculated for each fragment. The dataset was then divided into a training set (n=40) and a test set (n=10) using a sphere exclusion algorithm. Multiple linear regressions coupled with simulated annealing or stepwise regression resulted in model A (r2=0.92) and model B (r2=0.87), respectively. Leave-one-out validation showed that models A and B have internal predictive abilities of 72% and 61%, respectively. External validation indicated that both models are robust, with squared cross-correlation coefficients of the training set (pred-r2) of 0.87 and 0.89, respectively. The developed GQSAR models indicate that fragment R3 plays a key role in activity variation (65%) with sound contribution of five-membered rings (5 chain count), aromatic carbons (SaaaCE-index), and aromatic nitrogens (SaaNcount). In contrast, fragments R1 and R2 together contribute 35% of activity variation, suggesting that sulfur atoms (Sulfur count) and hydrophobic three-membered rings (chi3 chain) at R1 are preferable for inhibitory activity.
Descriptors; electrotopological indices; P70 ribosomal S6 kinase; quantitative structure activity relationship; regression analysis; simulated annealing, sphere exclusion.