PR002363 (Project)

Description:Untargeted metabolomics provides a sensitive readout of small molecules in biofluids, but requires targeted approaches to resolve ~90% of features for which tandem mass spectra (MS/MS) are not collected. By training on a subset of verified metabolites and their profiles in LC-MS, we derive a probabilistic model to predict molecular fingerprints in human stool and blood samples. These predictions, which do not utilize MS/MS, were accurate for >44% (correct top ranked candidate) or >75% (correct within top 3) of test metabolites, drastically reducing the number of reference standards that would need to be to be tested. These predictions revealed markers and drivers of inflammation, including amino acid derivatives and lysophospholipids with herein demonstrated platelet-activating factor receptor (PAF-R) activity. Integration with bacterial culturomics facilitates tracking the source of inflammation-associated metabolites to their origins in the gut microbiome.