Identification of oxidative stress-related diagnostic markers from gestational diabetes mellitus and immune infiltration by machine learning

Abstract

Abstract Background During pregnancy, the instability of glucose tolerance led to gestational diabetes mellitus (GDM) in the expectant mother, resulting in significant harm to both the pregnant woman and the infant. Oxidative stress-related genes (OSRGs) play a crucial role in diabetes. However, it is still not fully understood whether OSRGs also affect GDM. Therefore, this study sought to find hub genes from OSRGs can influence the diagnosis and treatment of GDM. Methods GSE203346 and GSE70493 from Gene Expression Omnibus (GEO) database were used as training and verification sets. OSRGs were downloaded from GeneCard database. The differentially expressed genes (DEGs) were identified with the limma package for GO/KEGG/GSEA analysis in the clusterProfiler package. These OSRGs were further analyzed using LASSO and SVM-RFE to identify diagnostic biomarkers for GDM. A diagnostic model was established using the Glm function and verified with ROC curve. The relationship between hub genes and immune infiltration was analyzed. Results The study identified 937 DEGs from GSE203346 dataset and 10 differentially expressed oxidative stress-related genes (DEOSRGs) were screened. The LASSO and SVM-RFE results showed that seven genes (DNMT3B, SLC6A2, DNMT1, CACNA1C, PENK, S100B and H2BC3) were identified as OSRGs diagnostic biomarkers for GDM. The AUC value of GDM diagnosis model was 0.9920 in training set and 0.8524 in verification set. Three hub genes (SLC6A2, PENK and S100B) were identified between DEOSRGs and LASSO and SVM-RFE results. ROC analysis showed that AUC values of SLC6A2, PENK and S100B were 0.7024, 0.7119 and 0.6964, respectively. The results of immune infiltration suggest that these SLC6A2, PENK and S100B might have regulatory role in the immune response of GDM. Finally, 19 drugs targeting S100B and SLC6A2 were retrieved from the DrugBank database, some of which may be useful for the treatment of GDM. Conclusions By combining DEOSRGs with two machine learning algorithms, three hub genes (SLC6A2, PENK and S100B) were explored. Three hub genes are strongly associated with several infiltrating immune cells, immunoinhibitors, immunostimulators and chemokines. These hub genes could potentially contribute to a better understanding of the development and drug treatment of GDM.