Identification of key immune genes of endometriosis based on bioinformatics and machine learning

Abstract

Abstract Introduction: Immunity and inflammation are involved in a multitude of reproductive metabolic processes, with a particular focus on endometriosis (EMT). The aim of this study is to employ bioinformatics methods to explore novel immune-related biomarkers and assess their predictive capabilities for EMT. Methods mRNA expression profiles were obtained from the GSE141549 and GSE7305 datasets in the Gene Expression Omnibus (GEO) database, while immune-related genes were sourced from the ImmPort database. Immune genes associated with EMT were filtered for differential analysis. Interrelationships between different immune-related genes (DIRGs) were characterized using protein-protein interaction (PPI) networks. Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses were applied to the functionality of DIRGs. Least Absolute Shrinkage and Selection Operation (LASSO) regression models and Boruta models were built to determine candidate genes for EMT, and the performance of the prediction models and candidate genes were verified using Receiver Operator Characterization curve (ROC) in GSE141549 and GSE7305. Finally, we structured the EMT prediction normogram on the basis of the five candidate DIRGs. Expression of the five candidate DIRGs in human samples was examined using PCR and Western Blot. The relative proportions of 22 immune cells were computed using the CIBERSORT algorithm, and the correlations between immune cells and candidate DIRGs were emphasized. Results Altogether 769 differentially expressed genes (DEGs) and 94 DIRGs were detected between ectopic and normal endometrium. These DIRGs were mainly concentrated in positive regulation of response to external stimulus, collagen-containing extracellular matrix, receptor ligand activity and signaling receptor activator activity. KEGG enrichment analysis mainly addressed Cytokine-cytokine receptor interaction and Neuroactive ligand-receptor interaction. Then, five key genes (SCG2, FOS, DES, GREM1, and PLA2G2A) were characterized using the GSE141549 dataset and used to build a prediction model for EMT. Conclusions Immunity and inflammation have a major role in the elaboration of EMT. SCG2, FOS, DES, GREM1 and PLA2G2A can serve as important biomarkers for EMT.