Characteristic gene prognostic model of type 1 diabetes mellitus via machine learning strategy

Abstract

The present study was designed to detect possible biomarkers associated with Type 1 diabetes mellitus (T1DM) incidence in an effort to develop novel treatments for this condition. Three mRNA expression datasets of peripheral blood mononuclear cells (PBMCs) were obtained from the GEO database. Differentially expressed genes (DEGs) between T1DM patients and healthy controls were identified by Limma package in R, and using the DEGs to conduct GO and DO pathway enrichment. The LASSO—SVM were used to screen the hub genes. We performed immune correlation analysis of hub genes and established a T1DM prognosis model. CIBERSORT algorithm was used to identify the different immune cells in distribution between T1DM and normal samples. The correlation of the hub genes and immune cells was analyzed by Spearman. ROC curves were used to assess the diagnostic value of genes in T1DM. A total of 60 immune related DEGs were obtained from the T1DM and normal samples. Then, DEGs were further screened to obtain 3 hub genes, ANP32A-IT1, ESCO2 and NBPF1. CIBERSORT analysis revealed the percentage of immune cells in each sample, indicating that there was significant difference in monocytes, T cells CD8+, gamma delta T cells, naive CD4+ T cells and activated memory CD4+ T cells between T1DM and normal samples. The area under curve (AUC) of ESCO2, ANP32A-IT1 and NBPF1 were all greater than 0.8, indicating that these three genes have high diagnostic value for T1DM. Together, the findings of these bioinformatics analyses thus identified key hub genes associated with T1DM development.