Prediction of the drug-specific CYP2C9 activity using deep learning method

Abstract

Pharmacogenomics is an area of how the individual responses to the drug based on the genomic information related to the drug response, which is directly linked to the personalized treatment. Many drugs are metabolized by cytochrome P450 (CYP450). Especially CYP2C9 metabolizes lots of drugs with narrow therapeutic index and has various activity by polymorphism, so the drug efficacy varies depending on the individual’s genotype. Due to the development of high-throughput sequencing, novel variants have been discovered continuously. Although the catalytic activity in CYP2C9 with decreased or no function is substrate-dependent, the current dosing guidelines are based on substrate-agnostic phenotypes, which are not consider the variability in drug response. However, it is time-consuming to identify the catalytic activity considering all combinations of drugs and alleles by biochemical experiments. Thus, the aim of this paper is to predict the drug-specific activity of CYP2C9 using deep learning method. The results show that the proposed model predicts the drug-specific CYP2C9 activity with a positive correlation with the real activity values. Especially, it shows better performance than traditional machine learning methods on the drug-specific activity prediction of unseen alleles. Therefore, it can provide the insightful understanding of the drug-specific activity of CYP2C9 variants.