Cytochrome P450 enzymes and in particular CYP 3A4 are responsible for the majority of phase I metabolism of drugs. Prediction of the most labile sites and the overall metabolic stability might help during the lead optimization process to achieve the desired pharmacokinetic drug candidate profile. Early work by Korzekwa et al. (JACS 1990, 11, 7042) was able to correlate the relative rates of hydroxylation with AM1 derived hydrogen abstraction activation energies. Singh et al. (J. Med. Chem. 2003, 46, 1330) established relative ranking of SOMs by hydrogen abstraction energy estimates through a QSAR model. By adding descriptors of accessibility, size and hydrophobicity, Sheridan et al. (J. Med. Chem. 2007, 50, 3137) generated random forest models which outperformed MetaSite (Cruciani et al. J. Med. Chem. 2005, 48, 6970). More recently SMARTCyp (Rydberg et al., ACS Med. Chem. Lett. 2012, 3, 69) achieved good SOM prediction based on substructure based reactivity rule-set derived from DFT calculations augmented with simple accessibility descriptors, while several other methods have been reported which rely more descriptor based machine learning (ML). In this work we investigate if QSAR/ML methods add value to physical rule based methods like SMARTCyp, and if performing QM calculations on the actual molecules improves the prediction. Datasets from the Sheridan paper and other public sources are used to compare the methods and external test sets are compiled from selected in house data. Finally we report on our efforts to build QSAR/ML models training on human liver microsomal stability and probe if employing the individual SOM scores as descriptors improves the models.

In silico approaches to CYP P450 site-of-metabolism (SOM) and microsomal stability prediction

249th National Meeting (2015)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.