Machine and deep learning identified metabolites and clinical features associated with gallstone disease

Abstract

Machine Learning (ML) algorithms can be used to analyze metabolomic expression data to explore the association between metabolite expression and disease etiology. In this study, we used and compared the performance of ML algorithms to analyze polar aqueous and blood-based lipid-based metabolites to identify meaningful patterns correlated with the development of gallstone disease (GSD) while examining the sex disparity. We also developed ML approaches that used clinical risk factors for predicting GSD, including age, obesity, body mass index, hemoglobin A1c, dyslipidemia index cholesterol to high-density lipoprotein ratio (CHOL/HDL). A more powerful data fusion model that combines both metabolomic and clinical features achieved accuracy of 83% for accurate prediction of the presence of GSD.