Unbiased “non-targeted” metabolite profiling techniques hold considerable promise for biomarker and pathway discovery, though there have been few successful applications to human disease. By integrating genomics, non-targeted metabolomics, and detailed human phenotyping we were able to structurally elucidate and identify dimethylguanidino valeric acid (DMGV) as a novel and independent marker of CT-defined nonalcoholic fatty liver disease (NAFLD) in Generation III of the Framingham Heart Study (FHS). After synthesis and purification, we developed a targeted mass spec assay whereby we demonstrated DMGV was also elevated in the plasma of hospital-based patients with biopsy-proven non-alcoholic steatohepatitis (NASH). DMGV levels fell in parallel with improvements in cardiometabolic parameters post gastric bypass surgery. In both the FHS and the Malmo Diet and Cancer Study, we also determined that DMGV was an independent predictor of future diabetes up to 12 years prior to disease onset. Finally, we provide all our metabolite peak data (known and unidentified peaks) and their associations with key metabolic parameters as a publicly available resource for researchers in metabolic disease.