QSAR analysis of diaryl COX-2 inhibitors: comparison of feature selection and train-test data selection methods.

Abstract

QSAR analyses were performed on a series of trans-stilbenoid diaryl compounds for modeling their COX-2 inhibitory activities. The multivariate regression equations were developed with the selected independent variables using various feature selection methods. In addition, model training was done using different test-train data selection methods. The applicability of each variable and the test-train selection methods was investigated through the type and number of the selected significant descriptors as well as the statistical criteria of the developed model for each pair of feature and test-train selection methods. The goodness of fit and the statistical significance of 15 developed equations were evaluated using the correlation coefficient (R), the variance ratio (F), and the standard error of estimate (S.E.). The models were validated using the leave many out and the leave one out cross-validation methods. The mean percentage deviation (MPD(+/-SD)) was used as an accuracy criterion for checking the predicted activities. It was found that the developed models could predict the COX-2 and COX-1 inhibitory activities as well as the COX-2/COX-1 selectivity ratios producing the MPD values of 1.6(+/-0.8)%, 7.7(+/-5.6)%, and 16.9(+/-9.6)%, respectively.