Prediction of Cytochrome P450 Mediated Metabolism using SMARTCyp and StarDrop

Patrik Rydberg, Optibrium Ltd

Just before going to press, we received the devastating news about Patrik Rydberg, who so kindly contributed an atricle about his SMARTCyp methodology for this newsletter a couple of weeks ago.  Prior to joining Optibrium, Patrik had a highly successful academic career at the University of Copenhagen where he undertook pioneering research into prediction of metabolism.  We are indebted to Patrik for his enormous contribution to the field of computational ADMET prediction, and our thoughts are with his family at this time.  As a mark of respect for Patrik and his work, we would like to dedicate this edition of the newsletter to his memory.

75% of all drugs are metabolized by enzymes belonging to the cytochrome P450 family. To predict where metabolism will occur on a compound during this process, there are several prediction models available, both free (e.g. SMARTCyp) and commercial (e.g. StarDrop).

In principle, to predict which sites on a molecule that are most likely to be oxidized by a P450 enzyme, we need to calculate the reactivity of each atom and the likelihood that the molecule will bind in an orientation that puts this atom close to the heme group at the bottom of the active site, where the chemical reactions take place.

Before joining Optibrium, I was employed at the University of Copenhagen where I developed SMARTCyp, a model depending only on 2D structures and fragment matching. This assigns reactivity to a potential site of metabolism by matching fragments against a library for which highly accurate transition state energies have been precomputed, using density functional theory; a site’s accessibility is determined using a few simple 2D descriptors and 2D pharmacophores. StarDrop, on the other hand, computes the reactivities on-the-fly using semiempirical AM1 calculations and uses more extensive descriptors to assess the accessibilities. While the advantage of SMARTCyp is its speed and robustness, the advantages of StarDrop are that substituent effects are properly handled and that changes within a chemical series can be followed by doing a calculation for each compound (SMARTCyp will often assign the same scores to atoms in similar compounds).

The current research at Optibrium involves taking the best parts of SMARTCyp and integrating these into StarDrop, as well as extending the P450 models in StarDrop to additional P450 isoforms. A common limitation of most metabolism prediction models is that they only predict which site will be oxidized and not what metabolite will be formed. While the rules for predicting the metabolites formed by metabolism at a site are straightforward in the majority of cases, we are interested in exploring the mechanisms that give rise to less common products, particularly the formation of reactive metabolites. We are researching the possibility to extend StarDrop with such models, as well as methods for predicting which isoforms will contribute to the metabolism of a compound.

Comments are closed.