Precision biomarker discovery in hypertension through explainable AI and proteomics.

Area of Science:

• Cardiovascular Disease Research
• Proteomics and Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Hypertension is a significant global health issue, driving cardiovascular disease.
• Limited availability of reliable blood-based biomarkers for early hypertension detection.
• Need for innovative diagnostic tools to identify hypertension at its initial stages.

Purpose of the Study:

• To identify circulating proteins associated with stage 1 hypertension using plasma proteomics.
• To leverage explainable machine learning for biomarker discovery in hypertension.
• To validate potential protein biomarkers for early hypertension detection.

Main Methods:

• Analysis of plasma proteomic profiles from 778 participants (224 hypertension cases, 554 controls) from the Qatar Biobank.
• Differential protein expression analysis (1305 proteins) adjusted for age and sex.
• Application of CatBoost classifier and SHapley Additive exPlanations (SHAP) for model interpretation.

Main Results:

• Identification of 36 proteins significantly associated with stage 1 hypertension.
• Observed characteristic patterns: lower Renin, sRAGE, ghrelin, IL-1RAcP; higher TFPI, QORL1, HSP70, C5a.
• CatBoost model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.7985.
• Pathway analysis indicated involvement of oxidative stress and vascular function.

Conclusions:

• Renin, TFPI, sRAGE, QORL1, ghrelin, HSP70, IL-1RAcP, and C5a are proposed as candidate circulating biomarkers for hypertension.
• Explainable AI effectively translates proteomic data into interpretable biomarker candidates.
• Further validation in diverse cohorts is essential for clinical application.