CYP Substrate Classification

CYP Substrate Classification

ADMET Predictor's human P450 substrate classification package includes CYP*_Substr models for nine human CYPs: 1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1, and 3A4. They predict whether or not a given molecule is likely to be a substrate for Phase 1 metabolism by these enzymes.

The data sets for the CYP substrate classification models were drawn from the databases like DIDB (hosted by the University of Washington), and GOSTAR database and other public resources (Zaretzki; 2012, Rendic 2002; Sheridan 2007). The FDA application review files were scanned for all the recently approved small molecule drugs. We extracted the metabolite data (along with corresponding responsible CYP isoforms) for newly approved drugs in the last five years (2015-2020). Data from all sources have been carefully (re)curated by reference to the original research literature. A large number of errors in chemical structures, dominant tautomeric state and classification (among others) have been identified and corrected.

Though the primary literature sometimes specifically indicates that a given compound is not metabolized by a particular CYP isoform, this is generally not the case for database entries or other compilations. Most "non-substrates" were categorized as such because their microsomal metabolism had been studied in some detail without having generated an indication that they were substrates for the corresponding CYP. Indeed, identifying and recategorizing nominal "negatives" produced by that assumption was a major part of the curation effort.

Compounds were originally deemed "well-studied" because they had been reported as substrates for at least one CYP isoform. This implicit constraint biased the models produced towards over-generalization, with all large molecules classified as CYP3A4 substrates and all bases as substrates for CYP2D6. This shortcoming was addressed by supplementing the data set with large molecules and metabolic intermediates (e.g., simple sugars and enzyme cofactors) which are demonstrably not CYP substrates.

The final data sets consist of 1,583 molecules for CYP1A2; 862 molecules for CYP2A6; 967 molecules for CYP2B6; 1008 molecules for CYP2C8; 1,811 molecules for CYP2C9; 1,523 molecules for CYP2C19; 1,974 molecules for CYP2D6; 938 molecules for CYP2E1; and 2377 molecules for CYP3A4. The fraction categorized as substrates were 0.40, 0.23, 0.34, 0.38, 0.28, 0.34, 0.38, 0.32 and 0.72, respectively. The table below summarizes the performance of our CYPxxx_Substr models using several statistics: concordance, sensitivity, and specificity.