Abstract:

MicroRNAs (miRNAs) play an important role in the occurrence of non-obstructive azoospermia (NOA). Nevertheless, there is still a lack of research on the molecular mechanisms by which miRNAs regulate target genes to mediate NOA at present. In this study, we obtained NOA-related miRNA datasets from the GEO database and applied differential expression matrices combined with weighted correlation network analysis (WGCNA) and LASSO regression to identify four key miRNAs. Based on the miRDB database, the target genes of these miRNAs were predicted and intersected with the differentially expressed genes (DEGs) in the NOA transcriptome datasets. This intersection resulted in the identification of 18 DEGs. The spermatogenesis score model revealed a significant positive correlation between the overall expression level of these 18 DEGs and spermatogenesis scores, suggesting their potential involvement in NOA development. These 18 DEGs were subsequently incorporated into machine learning, leading to the identification of four hub genes with high diagnostic value: MGARP, FER1L5, SNX2, and PAPOLB. In the NOA mouse model, MGARP and SNX2 were upregulated, whereas FER1L5 and PAPOLB were downregulated, consistent with the expression trends observed in the NOA datasets. These findings indicate that MGARP, FER1L5, SNX2, and PAPOLB may serve as novel biomarkers for NOA, providing a theoretical and experimental foundations for elucidating its mechanisms and improving clinical diagnosis.

Key words: miRNA, non-obstructive azoospermia, spermatogenesis, machine learning, bioinformatics