A refined 3-dimensional QSAR of cytochrome P450 2C9: computational predictions of drug interactions.

A ligand-based model is reported that predicts the Ki values for cytochrome P450 2C9 (CYP2C9) inhibitors. This CoMFA model was used to predict the affinity of 14 structurally diverse compounds not in the training set and appears to be robust. The mean error of the predictions is 6 microM. The experimentally measured Ki values of the 14 compounds range from 0.1 to 48 microM. Leave-one-out cross-validated partial least-squares gives a q2 value of between 0.6 and 0.8 for the various models which indicates internal consistency. Random assignment of biological data to structure leads to negative q2 values. These models are useful in that they establish a pharmacophore for binding to CYP2C9 that can be tested with site-directed mutagenesis. These models can also be used to screen for potential drug interactions and to design compounds that will not bind to this enzyme with high affinity.