Fig. 1: Two AI/ML models to derive the 5 multi-organ MetBAGs.

a We identified organ-specific metabolites for the five human organ systems by mapping the 107 non-derived metabolites to their corresponding plasma proteins using a linear regression model. The most strongly associated protein for each metabolite was then linked to their organ-specific RNA expression profiles using data from the Human Protein Atlas (e.g., S_HDL_P → LCAT→liver: https://www.proteinatlas.org/ENSG00000213398-LCAT/tissue). Method and Supplementary Table 2 detail the metabolite-protein-organ annotations. An interactive graph visualization for this annotation is also accessible at https://labs-laboratory.com/medicine/metabolite_organ_annotation. b We evaluated age prediction performance using the mean absolute error (MAE) on independent test (ind. test) data, employing 2 AI/ML models across 5 organ systems with 107 metabolites. The “#“ symbol indicates the model that generalized best to the independent test data (i.e., smallest Cohen’s D). Data are presented as box plots showing the distribution of MAE across different groups and organs, with the median indicated by a horizontal line inside the box and the interquartile range (IQR) spanning the box from the first to the third quartile. c The scatter plot between the AI/ML-derived biological age and chronological age without applying the age bias correction²³. Data are presented with points colored by organ and a linear regression line (black) fit to the data. Pearson’s correlation coefficient (r) and corresponding P-values were computed to assess the strength and significance of the linear relationship, with two-sided p-values provided.