Explainable machine learning framework for biomarker discovery by combining biological age and frailty prediction

Area of Science:

• Gerontology
• Biomarkers
• Artificial Intelligence

Background:

• Biological age (BA) and frailty are crucial health indicators in aging research.
• Understanding the relationship between blood biomarkers and these aging metrics is essential.

Purpose of the Study:

• To develop a novel framework integrating BA and frailty machine learning (ML) predictors with eXplainable Artificial Intelligence (XAI).
• To identify key blood-based biomarkers of aging using this integrated approach.

Main Methods:

• Utilized data from Chinese adults (≥45 years) from the China Health and Retirement Longitudinal Study (CHARLS).
• Employed four tree-based ML algorithms to predict BA and frailty using 16 blood biomarkers.
• Applied SHapley Additive exPlanations (SHAP) for in-depth biomarker analysis.

Main Results:

• CatBoost and Gradient Boosting models showed superior performance for BA and frailty prediction, respectively.
• Initial ML importance highlighted cystatin C and glycated hemoglobin.
• SHAP analysis revealed cystatin C as the primary contributor for both BA and frailty predictors.

Conclusions:

• The developed XAI-integrated framework provides a robust method for identifying aging biomarkers.
• Cystatin C emerges as a significant and consistent predictor of both biological age and frailty.
• This approach offers a scalable tool for quantitative understanding of aging biomarkers.