Deciphering Shared Gene Signatures and Immune Infiltration Characteristics Between Gestational Diabetes Mellitus and Preeclampsia by Integrated Bioinformatics Analysis and Machine Learning.

Gestational diabetes mellitus (GDM) and preeclampsia (PE) are common and serious disorders of pregnancy that threaten maternal safety and perinatal outcomes. Generally, GDM is recognized as an independent risk factor for the development of preeclampsia, while a history of preeclampsia in primiparous women is also a risk factor for GDM in subsequent pregnancies. However, the intricate underlying mechanisms of GDM and PE remain elusive. This study developed a diagnostic prediction model for GDM and PE. It investigated the correlation between shared signature genes and immune infiltration characteristics, by employing bioinformatic analysis combined with a machine learning strategy. The microarray datasets GSE103552 and GSE74341 from the Gene Expression Omnibus (GEO) database were used to obtain differentially expressed genes (DEGs). Then, signature genes were identified from the common DEGs via the methods of random forest (RF) algorithms, and artificial neural network (ANN) models. Furthermore, the immune infiltration patterns associated with GDM and PE were explored and validated in the training and testing sets. Moreover, to uncover the molecular mechanisms involved, an mRNA-miRNA network of target genes was constructed, and potential therapeutic drugs for GDM and PE were explored by querying the Connectivity Map (CMap) database. We obtained 45 DEGs by intersecting upregulated and downregulated DEGs from the GSE103552 and GSE74341 datasets. The results of GO annotation indicated that these 45 DEGs were mainly enriched in the process of cell cycle, and KEGG enrichment analysis indicated significant associations with immune signal transduction pathways and immune-related infectious disease. Six signature genes, namely TRA2A, NPM3, PHF5A, SNORD1C, PLXNA3, and C14orf142, were determined by machine learning models, and a diagnostic prediction model for GDM and PE was constructed based on these key genes, validating the highest prediction in the testing set. Moreover, we found increased infiltration of iDCs and T cell co-inhibition in the GDM group, while neutrophil, Th2 cell, and HLA levels were found to have decreased significantly. The PE group showed a significant increase in mast cells. In addition, the identified key genes were found to have potential associations with various immunocytes, immune functions, and checkpoints in the training and testing sets. Then, a miRNA-gene network analysis predicted several key miRNAs-miR-204, miR-23abc, miR-9, miR-205, and miR-455-5p-that might play significant roles in regulating these DEGs. In addition, the research also identified four potential therapeutic compounds for GDM (prima-1-met, geranylgeraniol, MLN-8054, and LY-364947), along with other drugs (deferiprone, peucedanin, MPEP, and IWR-1-endo) that could be targeted for treating PE. In summary, this work identified six signature genes (TRA2A, NPM3, PHF5A, SNORD1C, PLXNA3, and C14orf142) as potential genetic biomarkers for the diagnostic prediction of GDM and PE. A diagnostic prediction model was constructed based on these key genes, demonstrating strong performance when validated with an independent dataset. Moreover, we investigated the similarities and differences between the two diseases in terms of immune infiltration landscape and analyzed the correlations between key genes and the immune infiltration landscape, which provided insights into the molecular mechanisms underlying the development of GDM and PE. This understanding could pave the way for breakthroughs in identifying new immunotherapeutic targets and strategies for disease prevention and treatment.