ATP1A2 and RAB31 serve as candidate key genes in diabetes associated breast cancer identified by comprehensive bioinformatics and machine learning

This study employed systematic bioinformatics analyses and three machine learning-based feature selection frameworks to explore candidate key genes underlying diabetes-associated breast cancer (BC). Gene expression profiles of breast cancer were retrieved from The Cancer Genome Atlas (TCGA) database, and diabetes-related expression data were obtained from the Gene Expression Omnibus (GEO) dataset GSE95849. Shared differentially expressed genes (DEGs) were screened using the limma package in R software. Biological functions and signaling pathways of these DEGs were annotated via Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses using the clusterProfiler package. Three machine learning-based feature selection approaches—LASSO regression, support vector machine recursive feature elimination (SVM-RFE), and random forest coupled with Boruta feature selection—were used to identify hub genes associated with diabetes-related breast cancer. The prognostic value of candidate genes was validated in both TCGA-BC and METABRIC-BC cohorts, and protein expression levels were further verified by immunohistochemistry in clinical breast cancer tissues. A total of 96 overlapping DEGs with fully concordant direction of regulation (69 upregulated and 27 downregulated) were identified for in-depth analyses. Nine hub genes linked to diabetes-associated breast cancer were screened out, including ATP1A2, DNAJC1, DPT, EFNB3, EGFLAM, RAB31, TPPP, VCAN, and ZNF467. Among them, DNAJC1, RAB31, VCAN, and ZNF467 were upregulated, whereas ATP1A2, DPT, EFNB3, EGFLAM, and TPPP were downregulated. Functional enrichment analysis indicated that cell division, endocrine resistance, and the cAMP signaling pathway were critically implicated in diabetes-associated breast cancer. Survival analysis in the TCGA-BC cohort revealed that high expression of ATP1A2, EFNB3, RAB31, and TPPP was significantly correlated with favorable overall survival. Similarly, in the METABRIC-BC cohort, elevated expression of ATP1A2, DPT, EFNB3, RAB31, VCAN, and ZNF467 predicted better overall survival. ATP1A2 and RAB31 were identified as the candidate core genes for further investigation. In clinical diabetes-associated breast cancer tissues, ATP1A2 was notably downregulated while RAB31 was upregulated, suggesting their potential roles in the progression of diabetes-related breast cancer. Our findings indicate that ATP1A2 and RAB31 may serve as candidate key genes in diabetes-associated breast cancer, which offers useful clues for future mechanistic research and the development of targeted therapeutic strategies for this specific subtype of breast cancer.