Precision biomarker discovery in hypertension through explainable AI and proteomics.
This study identifies potential circulating biomarkers for stage 1 hypertension using advanced machine learning techniques, but these findings require further validation.
Where it sits
this study against the rest of the thymosin alpha-1 corpusSummary and findings
This study aimed to identify circulating proteins associated with stage 1 hypertension using plasma proteomics and machine learning in a cohort from the Qatar Biobank. A total of 778 participants were analyzed, leading to the identification of 36 proteins significantly associated with hypertension. The study highlights the potential of using explainable AI in biomarker discovery.
Abstract
Hypertension is a major global health burden and a leading driver of cardiovascular disease, yet reliable blood-based biomarkers for early disease are still limited. We combined plasma proteomics with explainable machine learning to identify circulating proteins associated with stage 1 hypertension in the Qatar Biobank. Proteomic profiles from 778 participants (554 controls and 224 stage 1 hypertension cases) were analyzed; 1305 proteins were tested for differential expression with adjustment for age and sex, and top features were prioritized before training predictive models. Among the evaluated classifiers, CatBoost performed best (AUROC = 0.7985), and SHapley Additive exPlanations were used to interpret the model. We identified 36 proteins significantly associated with hypertension and observed a characteristic pattern featuring lower Renin, sRAGE, ghrelin, and IL-1RAcP, and higher TFPI, QORL1, HSP70, and C5a in hypertensive individuals. Pathway and network analyses implicated processes related to oxidative stress and vascular function. Together, these results demonstrate Renin, TFPI, sRAGE, QORL1, ghrelin, HSP70, IL-1RAcP, and C5a as candidate circulating biomarkers for hypertension and illustrate the value of explainable AI for translating proteomic signals into potentially clinically interpretable candidates, pending validation in independent and diverse cohorts.