Machine learning deciphers the significance of mitochondrial regulators on the diagnosis and subtype classification in non-alcoholic fatty liver disease

Non-alcoholic fatty liver disease (NAFLD) is a highly prevalent liver disease worldwide and lack of research on the diagnostic utility of mitochondrial regulators in NAFLD. Mitochondrial dysfunction plays a pivotal role in the development and progression of NAFLD, especially oxidative stress and acidity β-oxidative overload. Thus, we aimed to identify and validate a panel of mitochondrial gene expression biomarkers for detection of NAFLD.

The four MRGs, namely BCL2L11, NAGS, HDHD3, and RMND1, are novel potential biomarkers for diagnosing NAFLD. A diagnostic model constructed using the four MRGs may aid early diagnosis of NAFLD in clinics.

We selected the GSE89632 dataset and identified key mitochondrial regulators by intersecting DEGs, WGCNA modules, and MRGs. Classification of NAFLD subtypes based on these key mitochondrial regulatory factors was performed, and the pattern of immune system infiltration in different NAFLD subtypes were also investigated. RF, LASSO, and SVM-RFE were employed to identify possible diagnostic biomarkers from key mitochondrial regulatory factors and the predictive power was demonstrated through ROC curves. Finally, we validated these potential diagnostic biomarkers in human peripheral blood samples and a high-fat diet-induced NAFLD mouse model.

We identified 25 key regulators of mitochondria and two NAFLD subtypes with different immune infiltration patterns. Four potential diagnostic biomarkers (BCL2L11, NAGS, HDHD3, and RMND1) were screened by three machine learning methods thereby establishing the diagnostic model, which showed favorable predictive power and achieved significant clinical benefit at certain threshold probabilities. Then, through internal and external validation, we identified and confirmed that BCL2L11 was significantly downregulated in NAFLD, while the other three were significantly upregulated.