Exploring biomarkers of premature ovarian insufficiency based on oxford nanopore transcriptional profile and machine learning

Abstract

Premature ovarian insufficiency (POI) is a reproductive endocrine disorder characterized by infertility and perimenopausal syndrome, with a highly heterogeneous genetic etiology and its mechanism is not fully understood. Therefore, we utilized Oxford Nanopore Technology (ONT) for the first time to characterize the full-length transcript profile, and revealed biomarkers, pathway and molecular mechanisms for POI by bioinformatics analysis and machine learning. Ultimately, we identified 272 differentially expressed genes, 858 core genes, and 25 hub genes by analysis of differential expression, gene set enrichment, and protein-protein interactions. Seven candidate genes were identified based on the intersection features of the random forest and Boruta algorithm. qRT-PCR results indicated that COX5A, UQCRFS1, LCK, RPS2 and EIF5A exhibited consistent expression trends with sequencing data and have potential as biomarkers. Additionally, GSEA analysis revealed that the pathophysiology of POI is closely associated with inhibition of the PI3K-AKT pathway, oxidative phosphorylation and DNA damage repair, as well as activation of inflammatory and apoptotic pathways. Furthermore, we emphasize that downregulation of respiratory chain enzyme complex subunits and inhibition of oxidative phosphorylation pathways play crucial roles in the pathophysiology of POI. In conclusion, our utilization of long-read sequencing has refined the annotation information within the POI transcriptional profile. This valuable data provides novel insights for further exploration into molecular regulatory networks and potential biomarkers associated with POI.

Figure 1

Flow charts based on analysis of bioinformatics and machine learning. DEA: differential expression analysis, DEGs: differentially expressed genes, GSEA: gene set enrichment analysis, PPI: protein–protein interactions.

Figure 2

Overall transcript expression levels and differentially expressed transcripts (DETs) in the two groups. (a) CPM density distribution map. (b) CPM boxplot of each sample. (c) Volcano plot of DETs. (d) MA plot of DETs. (e) Cluster heatmap of DETs expression.

Figure 3

Identification of differential core genes. (a) GSEA. (b) Venn diagram of DEGs and core genes. (c) Flow chart for identification of candidate genes. (d) Identification of hub genes by PPI.

Figure 4

Screening of candidate biomarkers based on machine learning. (a) Top10 feature variables based on random forest. (b) Variable attribute classification based on Boruta algorithm. Green, red and yellow represented accepted, rejected and tentative attributes respectively.

Figure 5

Relevance, expression and enrichment analysis of Hub genes. (a) Clinical Correlation Heat Map, * suggested P < 0.05, ** suggested P < 0.01. (b) Circos map of candidate gene expression. (c) Functional enrichment analysis of KEGG for hub gene. (d) qRT-PCR detection of candidate genes. (e) The down-regulated core genes, namely NDUFV1, UQCRFS1 and COX5A, are subunits of mitochondrial respiratory chain complex I, complex III and complex IV.