# Decoding the hypoxia-exosome-immune triad in OSA: PRCP/UCHL1/BTG2-driven metabolic dysregulation revealed by interpretable machine learning

## TL;DR

This study identifies three exosome-related genes linked to immune and metabolic issues in sleep apnea, using machine learning to predict their role and suggest potential treatments.

## Contribution

Novel integration of transcriptomic data and interpretable machine learning to identify exosome-driven immune-metabolic dysregulation in OSA.

## Key findings

- PRCP, UCHL1, and BTG2 are central biomarkers in OSA immune-metabolic disruption.
- XGBoost model achieved high diagnostic accuracy (AUC = 0.968) for these biomarkers.
- Immune cell infiltration analysis linked gene expression to CD56 bright NK cells and Memory B cells.

## Abstract

Obstructive sleep apnea (OSA) is a prevalent disorder characterized by significant metabolic and immune dysregulation. This study aims to uncover exosome-related biomarkers implicated in immune-metabolic disturbances in OSA and explore their potential as diagnostic and therapeutic targets.

Transcriptomic data from two GEO datasets (GSE135917 and GSE38792) were integrated and analyzed using differential expression analysis via the limma package. Key biomarkers were identified using feature selection techniques including LASSO and Random Forest. Machine learning models, specifically XGBoost, were trained to evaluate biomarker performance, with model accuracy assessed by ROC curve analysis and AUC values. Immune cell infiltration was evaluated using single-sample Gene Set Enrichment Analysis (ssGSEA). Drug enrichment predictions were made through the Drug Signatures Database (DSigDB). Vivo and Vitro Experimental Validation on Multiple Independent cohorts.

Three exosome-related biomarkers—PRCP, UCHL1, and BTG2—were identified as central to OSA’s immune-metabolic dysregulation. XGBoost modeling demonstrated robust predictive power (AUC = 0.968). Immune analysis revealed significant correlations between gene expression and immune cell subsets, particularly CD56 bright natural killer cells and Memory B cells. Drug enrichment analysis identified potential therapeutic compounds, including Pentaphenate and Delphinidin, which target these biomarkers. OSA is associated with a reproducible transcriptional signature characterized by increased PRCP and UCHL1 expression and decreased BTG2 expression.

This study identifies PRCP, UCHL1, and BTG2 as key exosome-related biomarkers in OSA that regulate immune-metabolic disruption. By integrating transcriptomic data, machine learning, and immune analysis, we uncover an “exosome-immune” axis in OSA pathophysiology.

Flowchart illustrating a process in three sections: Data sorting, Building the model, and Immune Correlation Analysis. Data sorting involves obtaining transcriptional data and using statistical methods like LASSO and Random Forest. Building the model includes visualizing gene expression, using ROC curves, and XGBoost model construction with SHAP for interpretation. Immune analysis utilizes ssGSEA for immune cell scores and DSigDB for therapeutic drug identification through protein-drug interactions.