Comparative structural connectivity spectra analysis (CoSCoSA) models of steroids binding to the aromatase enzyme

A method that combines NMR spectral and structural information into a constructed three-dimensional (3D)-connectivity matrix is developed for modeling biological binding activity of small molecules. The 3D-connectivity matrix for a molecule is defined by associating the distances between all possible carbon-to-carbon connections with their assigned carbon NMR chemical shifts. In this project we selected from the total 3D-connectivity matrix a subset, the two-dimensional (2D) (13)C-(13)C COSY and a theoretical long range 2D (13)C-(13)C distance connectivity spectral plane. Patterns of (13)C chemical shifts observed at these two relative distances for 50 steroids were used to produce a mathematical relationship for the steroids' relative binding affinity (pK(i)) to the aromatase enzyme. We call this technique comparative structural connectivity spectra analysis (CoSCoSA) modeling. Using combinations of the 2D COSY and 2D long-range distance spectra as modeling parameters, we built four CoSCoSA models. One model was made from the 2D COSY spectra alone and another was developed using only the 2D long-range distance spectra. Then the COSY and long-distance spectra were combined in two different ways: starting with the combined principal components (PCs) from the separately calculated COSY and distance spectra or using the combined raw spectra (3D). The best CoSCoSA model was based on the combined PCs from COSY and distance spectra. This model had an r(2) of 0.96 and a leave-one-out cross-validation (q(2)) of 0.92. In general CoSCoSA modeling combines the quantum mechanical information inherent in NMR chemical shifts with internal molecular atom-to-atom distances to give a reliable and straightforward basis for predictive modeling. The technique has the flexibility and accuracy to outperform not only the cross-validated variance q(2) of previously published quantitative structure-activity relationships (QSAR) but also those obtained by related quantitative spectral data-activity relationships (QSDARs) lacking connectivity dimensions.