The thirty-one benchmark steroids revisited: Comparative Molecular Moment Analysis (CoMMA) with principal component regression

A quantitative structure activity relationship (QSAR) study has examined the predicted activities of the thirty-one steroids that had been investigated in the original CoMFA study as well as in numerous subsequent studies. It is shown that the Comparative Molecular Moment (CoMMA) descriptors correlate closely with the measured corticosteroid binding affinities of the training set of twenty-one molecules. Principal component regression (PCR) coefficients obtained for this set are then used to predict the activities of the test set of the ten molecules outside of the training set. The results emphasize two different aspects of QSAR analysis. First, that molecules not representative of training set molecules should not be included in any test set used for training set validation; and second, that validation should not rely on the training set activity mean or offset in the regression equation.