Metabolomic aging and mortality clocks based on ensemble learning identify systemic aging risk across the human lifespan in 18,476 individuals
Data files
Sep 01, 2026 version files 13.29 MB
-
nom_diseased_val_7768_56.csv
5.59 MB
-
nom_healthy_train_5252_56.csv
3.78 MB
-
nom_healthy_val_5456_56.csv
3.92 MB
-
README.md
2.45 KB
Abstract
Aging is characterized by progressive functional decline, yet metabolomic signatures across the human lifespan remain incompletely defined. Here, we profiled plasma metabolomes from 18,476 participants aged 1-96 years across six centers in China using nano-assisted laser desorption/ionization mass spectrometry (LDI-MS) and identified 56 representative age-associated metabolites. In a healthy cohort, we developed a stacking ensemble model to estimate metabolomic age (mAge) and quantified metabolomic age acceleration (mAA) as the deviation of mAge from chronological age (cAge). Higher mAA was associated with health deterioration and unfavorable lifestyle factors. During 5.5 years of follow-up, individuals with accelerated mAA showed increased risk of incident chronic diseases, such as cancer, hypertension, and diabetes, as well as increased all-cause mortality. In a diseased cohort, we further developed a mortality-linked metabolomic measure (age deviation) that quantifies individualized departures from age-expected mortality risk inferred from plasma metabolites. Age deviation was associated with composite clinical health measures and increased with multi-system disease involvement and higher multimorbidity burden. Together, these findings establish a scalable, lifespan-wide metabolomics framework that yields clinically interpretable metabolomic aging metrics and captures disease-related departures from age-expected risk.
Description
This dataset contains normalized plasma metabolomic profiles (56 age-associated metabolite features) from 18,476 participants aged 1-96 years, profiled by nano-assisted laser desorption/ionization mass spectrometry (LDI-MS). Data were used to develop metabolomic aging clocks and mortality risk models as described in the associated manuscript.
Files
- nom_healthy_train_5252_56.csv — Healthy training cohort (n = 5,252), used for model development
- nom_healthy_val_5456_56.csv — Healthy validation cohort (n = 5,456), used for model evaluation
- nom_diseased_val_7768_56.csv — Diseased validation cohort (n = 7,768), used for mortality risk analysis
Variables
| Column | Description |
|---|---|
| id | De-identified sample identifier |
| group | Cohort assignment (healthy_train, healthy_val, diseased_val) |
| age | Chronological age in years |
| Remaining 56 columns | Normalized metabolite intensities, named by m/z value (e.g., 100.1121) |
Data Processing
Raw LDI-MS spectra were preprocessed (smoothing, baseline correction, peak detection, alignment, feature filtering) and batch-corrected using pyCombat across analytical batches and collection centers. The 56 age-associated metabolite features were selected by consensus of Random Forest, Lasso regression, and mutual information methods. Values are log-transformed and z-score normalized intensities.
Instrument
Bruker Autoflex III MALDI TOF/TOF, positive-ion reflector mode, m/z range 100-1000 Da.
Sample Source
Plasma from peripheral venous blood collected after overnight fasting, stored at -80 °C.
Human subjects data
This dataset contains only de-identified data. All participant identifiers were replaced with arbitrary coded IDs. No personally identifiable information is included. This study was approved by the Ethics Committee of Renji Hospital, Shanghai Jiao Tong University School of Medicine, and all participants provided written informed consent