Advantage of 2D‐QSAR in the discovery of novel protein geranylgeranyltransferase inhibitor (GGTI) scaffolds

Prenyltransferases selectively modify CaaX motif proteins (including Rho, Rap1, Rac, and Cdc42) with an isoprene lipid critically important for localization, function, and degradation. Selective GGTI's have been reported in the literature; however, all publicly available inhibitors are derived from essentially only two chemical scaffolds. We have extended quantitative structure‐activity relationship (QSAR) modeling traditionally reserved for lead optimization towards virtual screening. A series of 48 GGTI's were analyzed by 2D‐QSAR models using k‐nearest neighbor (kNN), automated lazy learning (ALL), and the partial least square (PLS) method. Validated QSAR models were used to mine ∼9.5 million compounds from public chemical databases. These searches resulted in 79 consensus hits that featured several novel scaffolds. Comparing structural similarity of database hits to the training set indicate many of the QSAR derived GGTI's are structurally divergent from the training set. Seven of these compounds with divergent backbones and high predicted GGTI activity were tested in vitro and all were found to be bona fide and selective inhibitors. These data demonstrate that validated QSAR models can serve as reliable virtual screening tools, providing higher hit rates as compared to chemical similarity searches. This strategy can be extended to any chemical biological system. Supported by F32‐GM073420