Multi-omics models for hypertension prediction using blood pressure PRS and omics scores
In the authors' words
Background: Hypertension is a major modifiable risk factor for cardiovascular disease, yet the relative contributions of proteomic (ProtS), metabolomic (MetS), and polygenic risk scores (PRS) to incident hypertension risk remain incompletely characterized. Methods: We developed ProtS and MetS for systolic (SBP) and diastolic blood pressure (DBP), and hypertension using Elastic Net regression in the Multi-Ethnic Study of Atherosclerosis (MESA). We incorporated these scores with previously developed BP PRSs into 5 incremental XGBoost models, with external validation in the Jackson Heart Study (JHS). We further applied transfer learning to adapt the MESA-derived models to JHS. Elastic Net selected proteins were prioritized using Mendelian randomization, colocalization, and functional analyses. Results: In MESA, model discrimination improved with the addition of omics scores, with the combined model achieving an area under the curve (AUC) of 0.72 compared with 0.63 for the clinical baseline model. In JHS, ProtS showed the highest AUC of 0.68 and the smallest cross-cohort performance decline, compared to MetS and PRS. Incorporating measured SBP and DBP substantially improved discrimination, with the PRS+BP model achieving an AUC of 0.80 in MESA and the baseline+BP model achieving an AUC of 0.74 in JHS. Among JHS participants with normal baseline BP, ProtS provided the strongest discrimination among models without measured BP (AUC = 0.62), while PRS+BP model achieved an AUC of 0.69. Leaf-weight refreshing consistently improved calibration in JHS while preserving discrimination. Mendelian randomization identified 77 proteins associated with BP traits or hypertension, involved in vascular, inflammatory, renal, and natriuretic pathways. Conclusions: ProtS demonstrated superior predictive performance and cross-cohort transferability among the omics modalities evaluated, while measured BP remained the strongest predictors of incident hypertension. Proteomic and genetic information may provide complementary biological signals for identifying individuals at elevated hypertension risk, particularly before clinically elevated BP is detectable.
Appeared: Tuesday, November 17. medRxiv. Preprint, not yet peer-reviewed.