A 2D QSAR analysis with three descriptors of binding affinity to human cytosol receptor was performed. The set of twenty-three progestins was divided into a training set of sixteen molecules and a test set of seven molecules. The quantum chemical RM1 semiempirical method was used to calculate geometry and some molecular properties. DRAGON software was also use to produce descriptors. MobyDigs software was used to select descriptors and build QSAR models. The best 2D QSAR model was constructed for the training set with multiple linear regression (MLR) using three descriptors , PW2, Mor15m, and GAP-10, resulting in r² = 0.866, q² = 0.805, q²boot = 0.723, q²ext = 0.666. A set of nine additional progestins that were not used for model building was used for external validation resulting q²ext = 0.403. The QSAR model was also validated by RQK fitness functions. It was shown to satisfy all the required criteria for validation. Two 3D QSAR models were built, first, to estimate predictive power, second, to analyze it. The predictive power of the 3D QSAR obtained with the nine external validation compounds was q²ext = 0.476. Based upon the graphical representation of PLS regression coefficients corresponding to steric and electrostatic interactions, it was possible to obtain a mechanistic interpretation. Thus the 2D and 3D QSAR together satisfy all the six Setubal Principles (OECD principles). Based upon the information obtained from the 3D QSAR analysis, the structures of four new progestins are proposed. Their receptor binding activities are estimated to be several times more potent than the most potent progestin of the twenty-three studied.
progestins; relative binding affinity; 2D and 3D QSAR
