Beiying Deng,1,2,* Ying Chen,1,2,* Pengzhan He,1,2,* Yinghui Liu,3 Yangbo Li,1,2 Yuli Cai,4 Weiguo Dong1 

1Department of Gastroenterology, Renmin Hospital of Wuhan University, Wuhan, People’s Republic of China; 2Central Laboratory, Renmin Hospital of Wuhan University, Wuhan, People’s Republic of China; 3Department of Geriatric, Renmin Hospital of Wuhan University, Wuhan, People’s Republic of China; 4Department of Endocrinology, Renmin Hospital of Wuhan University, Wuhan, People’s Republic of China

*These authors contributed equally to this work

Correspondence: Weiguo Dong, Department of Gastroenterology, Renmin Hospital of Wuhan University, No. 99 Zhangzhidong Road, Wuhan, Hubei Province, 430060, People’s Republic of China, Email dongweiguo@whu.edu.cn

Background: This study aims to elucidate the role of mitochondrial autophagy in metabolic dysfunction-associated steatohepatitis (MASH) by identifying and validating key mitophagy-related genes and diagnostic models with diagnostic potential.
Methods: The gene expression profiles and clinical information of MASH patients and healthy controls were obtained from the Gene Expression Omnibus database (GEO). Limma and functional enrichment analysis were used to identify the mitophagy-related differentially expressed genes (mito-DEGs) in MASH patients. Machine learning models were used to select key mito-DEGs and evaluate their efficacy in the early diagnosis of MASH. The expression levels of the key mito-DEGs were validated using datasets and cell models. A nomogram was constructed to assess the risk of MASH progression based on the expression of the key mito-DEGs. The mitophagy-related molecular subtypes of MASH were evaluated.
Results: Four mito-DEGs, namely MRAS, RAB7B, RETREG1, and TIGAR were identified. Among the machine learning models employed, the Support Vector Machine demonstrated the highest AUC value of 0.935, while the Light Gradient Boosting model exhibited the highest accuracy (0.9189), kappa (0.7204), and F1-score (0.9508) values. Based on these models, MRAS, RAB7B, and RETREG1 were selected for further analysis. The logistic regression model based on these genes could accurately predict MASH diagnosis. The nomogram model based on these DEGs exhibited excellent prediction performance. The expression levels of the three mito-DEGs were validated in the independent datasets and cell models, and the results were found to be consistent with the findings obtained through bioinformatics analysis. Furthermore, our findings revealed significant differences in gene expression patterns, immune characteristics, biological functions, and enrichment pathways between the mitophagy-related molecular subtypes of MASH. Subtype-specific small-molecule drugs were identified using the CMap database.
Conclusion: Our research provides novel insights into the role of mitophagy in MASH and uncovers novel targets for predictive and personalized MASH treatments.

Keywords: metabolic dysfunction-associated steatohepatitis, mitophagy, biomarkers, diagnostic model, machine learning