2D-QSAR Autocorrelation Study on Selective COX-2 Inhibitors

Motivation. A database of 185 COX–2 inhibitors including traditional NSAIDs was used to derive regression and classification models starting from the 2D autocorrelation descriptors. For that, we have described an original application of autocorrelation vectors for encoding the chemical information derived from 2D– structures. Method. Thus, a separate autocorrelation vector was computed for each of the following atomic properties: Pauling electronegativity, van der Waals volume, indicator for saturated/unsaturated, logP contribution, polar surface area, indicator for hydrogen–bond donor, indicator for hydrogen–bond acceptor and partial charge, and the resulting set was reduced into a smaller number of variables using PCA. Results. The robustness and predictive ability of the obtained models were evaluated by selecting chemical classes similar to the training set with activities ranging from low to high and the results of the models were compared. Conclusions. The data generated from the present study should be useful for predictive purposes, as an in silico filter, to screen large structural database for new potent COX–2 inhibitor.