Identification of mitochondria-related feature genes for predicting type 2 diabetes mellitus using machine learning...

We aimed to identify the mitochondria-related feature genes associated with type 2 diabetes mellitus and explore their potential roles in immune cell infiltration.

Datasets from GSE41762, GSE38642, GSE25724, and GSE20966 were obtained from the Gene Expression Omnibus database. Weighted Gene Co-expression Network Analysis was performed to achieve mitochondria-related hub genes. Random Forest, Least Absolute Shrinkage and Selection Operator, and Support Vector Machines-Recursive Feature Elimination algorithms were used to screen mitochondria-related feature genes. Receiver Operating Characteristic analysis was applied to evaluate the accuracy of the feature genes. Pearson's correlation analysis was used to calculate the correlations between feature genes and immune cell infiltration. The prediction of candidate drugs targeting the feature genes were predicted using the DGIdb database. qRT-PCR was performed to access the mRNA expressions of the feature genes.

Five mitochondria-related feature genes (SLC2A2, ENTPD3, ARG2, CHL1, and RASGRP1) were identified for type 2 diabetes mellitus prediction. They possessed high predictive accuracies with the area under the Receiver Operating Characteristic curve values >0.8. All five genes showed the strongest positive correlation with regulatory T cells and negative correlation with neutrophils. Additionally, drugs prediction analysis revealed 2(S)-amino-6-boronohexanoic acid, difluoromethylornithine, and compound 9 could target ARG2, while metformin was a candidate drug for SCL2A2. Finally, all five genes were confirmed to be decreased in MIN6 cells treated with high glucose and palmitic acid.

SLC2A2, ENTPD3, ARG2, CHL1, and RASGRP1 could be used as the mitochondria-related feature genes to predict type 2 diabetes mellitus and the therapeutic targets.