ProtPhenoAge: Integrating plasma proteomics to predict Aging-Related disease Risks
Journal of Advanced Research, 2026
Wang Y., Sun Y., Yang F., Li M., Qi T., Lu Z., Wang Q., Bu Q., Sun L., Wo H., Zhao Y., Yi H., Dai J.
| Disease area | Application area | Sample type | Products |
|---|---|---|---|
Aging | Pathophysiology Patient Stratification | Plasma |  Olink Explore 3072/384 |
Abstract
Introduction
Plasma proteins reflect the combined influence of both internal and external factors, making proteomics-based aging clocks a promising approach for quantifying the aging process.
Objective
This study aims to develop and validate a novel proteomics-based aging clock by integrating plasma proteomics with composite biomarkers.
Methods
We used a prospective cohort of 37,433 participants (median follow-up: 164.73 months) from the UK Biobank (UKB) with Olink Explore data. We calculated biological age (PhenoAge) and used the Boruta-SHAP (SHapley Additive exPlanations) algorithm to select PhenoAge-related proteins. Based on these proteins, six machine learning models were trained to develop a proteomics-based PhenoAge (ProtPhenoAge). We selected the best model as ProtPhenoAge based on the predictive capabilities of each model for PhenoAge and all-cause mortality. Phenome-wide association study (PheWAS) and Mendelian randomization (MR) explored associations between ProtPhenoAge Acceleration (ProtPhenoAgeAccel) and phenotypes. Genome-wide association study (GWAS) and colocalization analysis identified aging-associated loci.
Results
A total of 185 PhenoAge-related plasma proteins were used to develop ProtPhenoAge. The ProtPhenoAge model, using extreme gradient boosting (XGBoost), showed strong correlation with PhenoAge (r = 0.96, R2 = 0.92) and performed well in predicting all-cause mortality [area under the curve (AUC) = 0.76], outperforming previous aging clocks: chronological age (CA), PhenoAge and ProtAge. ProtPhenoAgeAccel was significantly associated with 313 disease phenotypes, covering a broad range of aging-related phenotypes. Compared with previous clocks, it identified more age-independent but aging-related phenotypes. In the GWAS, we identified 10 aging-associated loci. Among them, rs1045929 (P = 2.61 × 10-9) and rs429358 (P = 7.96 × 10-12) are respectively related to epigenetic aging and the well-recognized aging gene APOE.
Conclusion
Based on genomic and phenomic evidences, ProtPhenoAge was regarded to better quantifies the aging process by overcoming the limitations of previous clocks, which failed to detect time-independent aging features. These findings suggested that ProtPhenoAge is a reliable tool to assess aging and supporting aging research.