Revealing therapeutically actionable targets in Obstructive Sleep Apnea through machine learning-based miRNome analysis

Background: Obstructive sleep apnea (OSA) is associated with significant morbidity and mortality. Understanding its multiple pathogenetic pathways is crucial for improving patient outcomes. MicroRNA (miRNA) profiling coupled with machine learning (ML) approaches constitute a promise strategy for elucidating molecular pathways and identifying therapeutic targets.

Objective: To identify OSA-related mechanistic pathways and potential therapeutic targets utilizing miRNA profiling, ML and bioinformatics.

Methods: Multicenter study including 526 patients with suspected OSA. Plasma miRNAs were quantified using miRNA sequencing and RT-qPCR. ML methods and bioinformatics analyses were employed. Further analyses included publicly available datasets.

Results: Forty-two miRNAs showed differences between OSA and non-OSA patients in the miRNA sequencing phase (n=53). Eleven miRNAs were validated by RT-qPCR (n=53) and quantified in the remaining population (n=473). Among these, let-7d-5p, miR-15a-5p, and miR-107 were selected by three ML methods (Boruta, SPLS and VSURF). Enrichment analyses (KEGG, GO and Reactome) identified mechanisms linked to cell death, metabolism and fibrosis, among others. Two target genes (ATG9A and TFDP2) exhibited differences between OSA and non-OSA patients in three transcriptomic datasets. Continuous positive airway pressure (CPAP) treatment in OSA patients did not alter the levels of both targets in two additional datasets. No FDA-approved drugs were identified in Drug-Gene Interaction analysis.

Conclusions: The identified miRNAs and their target genes provide novel insights into the pathogenesis of OSA and would reveal therapeutic opportunities.