Classification models for CYP450 3A4 inhibitors and non-inhibitors

Cytochrome P450 3A4 (CYP3A4) is the predominant enzyme involved in the oxidative metabolic pathways of many drugs. The inhibition of this enzyme in many cases leads to an undesired accumulation of the administered therapeutic agent. The purpose of this study is to develop in silico model that can effectively distinguish human CYP3A4 inhibitors from non-inhibitors. Structural diversity of the drug-like compounds CYP3A4 inhibitors and non-inhibitors was obtained from Fujitsu Database and Korea Research Institute of Chemical Technology (KRICT) as training and test sets, respectively. Recursive Partitioning (RP) method was introduced for the classification of inhibitor and non-inhibitor of CYP3A4 because it is an easy and quick classification method to implement. The 2D molecular descriptors were used to classify the compounds into respective inhibitors and non-inhibitors by calculation of the physicochemical properties of CYP3A4 inhibitors such as molecular weights and fractions of 2D VSA chargeable groups. The RP tree model reached 72.33% of accuracy and exceeded this percentage for the sensitivity (75.82%) parameter. This model is further validated by the test set where both accuracy and sensitivity were 72.58% and 82.64%, respectively. The accuracy of the random forest model was increased to 73.8%. The 2D descriptors sufficiently represented the molecular features of CYP3A4 inhibitors. Our model can be used for the prediction of either CYP3A4 inhibitors or non-inhibitors in the early stages of the drug discovery process.

The Recursive Partitioning decision tree by PreADMET molecular descriptors to classify two active classes, inhibitors and non-inhibitors of CYP3A4, has been generated.Figure optionsDownload as PowerPoint slide