A chemoinformatics approach for assessing genetic missense variants with potential implications in Cytochrome P450 enzymes.

Sitabule, B. R.*, Hazelhurst, S., Othman, H.

Sydney Brenner Institute for Molecular Bioscience, Human Genetics, University of the Witwatersrand

Cytochrome P450 (CYP P450) enzymes are involved in over 90% of metabolic reactions that are known; variants in these enzymes result in varying drug responses among patients. To characterize genetic variants that affect drug response, a variety of experiments have been performed including in silico methods such as machine learning (ML) predictions. ML models can predict the potential impact of variants on enzyme interactions with ligands. We will develop an ML model that predicts the potential effect of missense variants on CYP P450 enzyme and ligand interactions by collecting and analyzing over 30 000 enzyme assay data from BindingDB. To train the ML model, CYP P450 enzymes were encoded using AAindex indices and the ligands were encoded using molecular descriptors, which were generated by RDkit and Mordred. In addition, the IC50 values of the enzyme assays were also encoded for the training of the model. The ML model will be trained, and its performance will be assessed. For this study, it is expected that a model which predicts the potential impact of missense variants on the interactions of CYP P450 enzymes with ligands will be developed. This may help characterize CYP P450 enzyme missense variants and gain insight on how missense variants affect enzyme activity, which in turn will contribute to the development of precision medicine.