Machine learning identification of biomarkers for vascular endothelial damage in obstructive sleep apnea

Obstructive Sleep Apnea (OSA) is a common sleep disorder that, if left untreated, may lead to chronic damage to multiple organs. The purpose of this study is to use machine learning methods to identify biomarkers associated with Vascular Endothelial Damage (VED), providing insights for early diagnosis and targeted therapy of OSA.

The authors conducted a differential expression analysis on the dataset GSE75097 to obtain Differentially Expressed Genes (DEGs), and obtained 1577 VED-Related Genes (VEDRGs) from the GeneCards database. The intersection genes of the two were defined as the differentially expressed VED-Related Genes (DE-VEDRGs). Functional and pathway enrichment analyses were conducted on them. Then, the authors adopted a combination of three machine learning algorithms to screen the Feature DE-VEDRGs (FDE-VEDRGs) of OSA. Furthermore, the authors conducted differential, correlation, and ROC curve analyses on FDE-VEDRGs, and performed Gene Set Enrichment Analysis (GSEA) and Gene Set Variation Analysis (GSVA) on these genes. Ultimately, an immune cell infiltration analysis was conducted between OSA and Primarily Snore (PS) samples, and the relationship between the expressions of FDE-VEDRGs and the levels of immune cell infiltrations was evaluated.

A total of 40 DE-VEDRGs were identified. These genes were involved in biological functions such as wound healing, small GTPase-mediated signal transduction, collagen-containing extracellular matrix, apical part of the cell, enzyme inhibitor activity, and growth factor receptor binding. The main pathways included Proteoglycans in cancer, Lipid and atherosclerosis, MicroRNAs in cancer, AGE-RAGE and PI3K-Akt signaling pathways, Apoptosis, etc. Six FDE-VEDRGs of OSA, namely F2RL2, SLC12A1, TGFB2, SPP1, PLAUR and MMP3, were identified, and their expression differences and correlations were explained. Furthermore, GSEA and GSVA analyses revealed significant differences in function and pathway enrichment between the high- and low- expression groups of MMP3, PLAUR, and SPP1. Finally, the analysis of immune cell infiltration indicated that compared with the PS group, the proportion of NK cell activation and macrophage M2 in the OSA group was significantly decreased, while the proportion of neutrophils was significantly increased. The correlation matrix shows the correlation between each FDE-VEDRG and each type of immune-infiltrating cells. The six-gene logistic model showed moderate discriminative ability in the discovery dataset (AUC = 0.78) and consistent performance in two external datasets (AUC = 0.75 and 0.72), suggesting potential predictive value that warrants further validation.

This study identified several candidate vascular endothelial damage-related genes in OSA using integrative bioinformatics and machine learning analyses. These findings may provide potential biomarkers and preliminary mechanistic clues for future experimental and translational studies on OSA.