Benchmarking analysis of deleterious SNP prediction tools on CYP2D6 enzyme

Abstract

The cytochrome P450 family is composed of hemeproteins involved in the metabolic transformation of endogenous and exogenous substances. The CYP2D6 enzyme is responsible for the metabolism of approximately 25% of clinically used drugs and is mainly expressed in the liver. The CYP2D6 gene is known to have a large number of Single Nucleotide Polymorphisms (SNPs) and the majority of them do not present clinical consequences. Nevertheless, these variations could modify the CYP2D6 enzymes function, resulting in poor metabolizing or ultra-extensive metabolizing phenotypes, when metabolism is slower or accelerated, respectively. Currently, there are several computational tools for predicting functional changes caused by genetic variations. Here, using 20 web servers, we evaluated the impact of 21 missense SNPs (6 neutral and 15 deleterious) previously validated by the literature. Only seven predictors presented sensitivity higher than 70%, while four showed specificity higher than 70% and only one reached the Matthews correlation coefficient of 0.39. Combinations of tools with greater sensitivity and specificity were made to improve the Matthews correlation coefficient, which increased the coefficient of five tools (Provean, FatHMM, SDM, PoPMuSiC and HotMuSiC). The results suggest that the most appropriate tool for CYP2D6 SNP prediction is FATHMM, which could aid in the classification of novel missense SNPs in this gene, providing the identification of mutations potentially associated with drug metabolism.