Raw metabolite integrals were normalized to the median value (median value = 1) of each 
metabolite for a given run day. Metabolite identification was performed using automated 
comparison to features (including retention time, m/z, preferred adducts, in-source fragments, 
and mass spectra) of ~8,000 purified chemical standards in the Metabolon reference library 
(Dehaven CD, Evans AM, Dai H, Lawton KA. Organization of GC/MS and LC/MS 
metabolomics data into chemical libraries. Journal of Cheminformatics. 2010;2(1):1-12.). 
Metabolites that remained unknown, but are reproducibly captured, after comparison with 
standards were annotated with a unique and persistent “X-number” for potential future 
identification.  
Abundances for a total of 1,546 grouped and aligned metabolites were reported by the 
Metabolon platform. Metabolites annotated as xenobiotics and with ≥ 75% missing values were 
removed, yielding 1,108 metabolites in the final dataset. Remaining missing values were 
imputed as the half-minimum intensity for a given metabolite. Metabolite intensities were log-
transformed and pareto-scaled for use in regression models.   
 
