DNA-methylation based Machine-learning Model for Impaired Liver Function Prediction

Abstract

Liver diseases include any medical disorders that negatively impact the normal functions of liver. They pose a great threat to public health due to their high prevalence. The early diagnosis of liver diseases is important for successful treatment but is hindered by the fact that liver diseases tend to show few symptoms in their early stages. In this project, I collected DNA methylation data from normal liver samples (n = 191) as well as data generated from early and late stage of liver diseases (n = 756) and identified 258 loci that were differentially methylated between all studied disease groups and the normal group. Using these CpGs as features, I trained a predictive SVM (support vector machine) model to discriminate whether a liver is diseased, and then used the 10-fold cross-validations to evaluate the model's predictive skills. The SVM model achieved outstanding classification power for impaired and normal livers, with AUROC (Area Under the Receiver Operating Characteristic curve) = 0.95, precision = 0.93, and recall = 0.96. The DNA methylation markers discovered in this study promise early diagnosis of liver diseases and pave the way for the future development of preventive and therapeutic epigenetic agents.