Prediction of cytochrome P450 mediated metabolism

Patrik Rydberg, University of Copenhagen, Denmark

The oxidation of compounds by the cytochrome P450 (CYP) enzyme family can lead to products with properties that have significantly different properties compared to the original compounds. Hence, to predict toxicity the metabolites that are most likely to be formed through reactions performed by CYPs should be taken into account.

To enable automatic inclusions of metabolites in toxicity predictions we have created the SMARTCyp methodology which predicts the most likely site of metabolism in CYPs. SMARTCyp gives the correct site of metabolism in ~80% drug-like molecules, which is comparable to, or better than, other methods for site of metabolism prediction.

SMARTCyp is the first step towards a method that not only predicts the correct site of metabolism, but also is correct for the right reasons. The method has been built from reactivity data that has been computed using state-of-the-art quantum chemical methods. The calculations have been performed with a model of a heme group (the co-factor which performs the reactions in CYPs and is common to all CYPs), to be CYP isoform unspecific, enabling isoform specific models to be built by combining SMARTCyp with external isoform specific data. So far, we have built isoform specific methods for nine CYP isoforms using machine learning and additional descriptors, resulting in performances of 82-87% for all isoforms on large test sets.

The SMARTCyp method itself is very transparent, enabling the user to understand why certain sites are predicted to be metabolized. Since the method is based on fragment based reactivities the computation time is very fast. SMARTCyp has been implemented using the open source CDK java library to enable anyone to implement the method into their own software and workflows.