Identification of exosomal miRNA-based predictive signatures for gestational diabetes mellitus via multi-algorithm...

Area of Science:

• Biomarkers
• Genomics
• Metabolic Disorders

Background:

• Gestational diabetes mellitus (GDM) is a common pregnancy complication with adverse outcomes.
• Exosomal microRNAs (exo-miRNAs) are promising noninvasive biomarkers for GDM.
• Current diagnostic models for early GDM prediction using exo-miRNAs are insufficient.

Purpose of the Study:

• To identify differentially expressed exo-miRNAs (DE-exo-miRNAs) for GDM prediction.
• To develop and validate a robust machine learning (ML) model for early GDM detection.
• To assess the generalizability of the predictive model across independent cohorts.

Main Methods:

• Utilized GSE192813 dataset to identify DE-exo-miRNAs between GDM and normal glucose tolerance (NGT) pregnancies.
• Applied five ML feature selection algorithms (LASSO, Random Forest, SVM-RFE, XGBoost, Boruta) to identify predictive exo-miRNAs.
• Developed 50 distinct ML models by combining feature selection with ten classification algorithms, and validated the best model using GSE114860.

Main Results:

• Identified 12 DE-miRNAs, with miR-423-5p, miR-99a-5p, miR-148a-3p, miR-192-5p, and miR-122-5p selected by multiple algorithms.
• The XGBoost + Boruta model achieved >90% accuracy (AUC >0.90) in the discovery cohort and >80% accuracy in external validation.
• Functional analysis revealed target genes involved in insulin signaling, lipid metabolism, and inflammatory pathways.

Conclusions:

• An integrative ML framework successfully identified a robust exo-miRNAs-based predictive signature for GDM.
• The developed model demonstrates high diagnostic accuracy and generalizability for early GDM screening.
• This approach holds potential for noninvasive early detection and precision management of gestational diabetes mellitus.