About the Prediction Method
How the metabolic age predictor works, its validation performance, and preprocessing pipeline.
Leave-One-Out Cross-Validation: Predicted vs Chronological Age
Grey dots = training samples (n=31) under LOOCV.
Dashed line = perfect prediction.
Note: example patient files have their chronological age (AGE_CALC) embedded in the CSV,
so a comparison is shown automatically when you use the example buttons.
Model Performance (Leave-One-Out CV)
4.58
MAE (years)
0.370
R²
0.654
Pearson r
0.659
Spearman ρ
Method & Preprocessing
- Algorithm
- BAYESIANRIDGE regression
- Preprocessing
- Robust scaling (median / IQR), median imputation for below-LOD values
- Feature engineering
- Raw metabolite features (no engineered features)
- Feature selection
- PASSTHROUGH (top 75 features)
- Validation
- Leave-One-Out CV on 31 samples
- Trained
- 2026-05-06