BACKGROUND: The global prevalence of non-alcoholic fatty liver disease (NAFLD) is approximately 30%, and the condition can progress to non-alcoholic steatohepatitis, cirrhosis, and hepatocellular carcinoma. Metabolic and bariatric surgery (MBS) has been shown to be effective in treating obesity and related disorders, including NAFLD. OBJECTIVE: In this study, comprehensive machine learning was used to identify biomarkers for precise treatment of NAFLD from the perspective of MBS. METHODS: Differential expression and univariate logistic regression analyses were performed on lipid metabolism-related genes in a training dataset (GSE83452) and two validation datasets (GSE106737 and GSE48452) to identify consensus-predicted genes (CPGs). Subsequently, 13 machine learning algorithms were integrated into 99 combinations
among which the optimal combination was selected based on the total score of the area under the curve, accuracy, F-score, and recall in the two validation datasets. Hub genes were selected based on their importance ranking in the algorithms and the frequency of their occurrence. Finally, a mouse model of MBS was established, and the mRNA expression of the hub genes was validated via quantitative PCR. RESULTS: A total of 12 CPGs were identified after intersecting the results of differential expression and logistic regression analyses on a Venn diagram. Four machine learning algorithms with the highest total scores were identified as optimal models. Additionally, PPARA, PLIN2, MED13, INSIG1, CPT1A, and ALOX5AP were identified as hub genes. The mRNA expression patterns of these genes in mice subjected to MBS were consistent with those observed in the three datasets. CONCLUSION: Altogether, the six hub genes identified in this study are important for the treatment of NAFLD via MBS and hold substantial promise in guiding personalized treatment of NAFLD in clinical settings.