Machine learning-based identification of lactate metabolism-associated biomarkers in non-alcoholic fatty liver disease.

Abstract

To determine lactate metabolism-associated biomarkers for non-alcoholic fatty liver disease (NAFLD). Based on NAFLD datasets from the gene expression omnibus database and lactate metabolism-related genes from GeneCards database, NAFLD-lactate metabolism hub genes (NAFLD-LM HGs) were screened via differential analysis, weighted gene co-expression network analysis and four machine learning algorithms. Their diagnostic efficacy was evaluated; molecular subtyping was performed; single-cell RNA sequencing (scRNA-seq) was subsequently conducted; and validation was finally conducted in NAFLD mouse models. We finally screened out eight NAFLD-LM HGs, namely CCAAT/enhancer-binding protein alpha (CEBPA), flavin containing dimethylaniline monooxygenase 1 (FMO1), insulin-like growth factor-binding protein 1 (IGFBP1), krüppel-like factor 4 (KLF4), low-density lipoprotein receptor (LDLR), myelocytomatosis oncogene (MYC), nuclear receptor subfamily 4 group A member 2 (NR4A2), and thymidylate synthase (TYMS), with a diagnostic rate of 0.798 and an area under the curve of 0.948. These genes drove the molecular heterogeneity in NAFLD patients by modulating of metabolism and immune microenvironment. The results of scRNA-seq clarified that Th17 cells were identified as the most abundant annotated cell subset. Through animal experiments, five genes (LDLR, MYC, IGFBP1, NR4A2, and KLF4) were established to have potential protective effects against NAFLD. Five genes (LDLR, MYC, IGFBP1, NR4A2, KLF4) are identified as "protective factors", and their downregulation is associated with NAFLD progression. The remaining three genes (CEBPA, FMO1, TYMS) exhibit inconsistent expression patterns, suggesting bidirectional regulation.