Assessing the utility of metabolite profiles of human serum, urine and saliva for determining dietary intake

This same metabolomics data was then used to model two dietary scores measuring three key aspects of complex diets: dietary composition, grouped by k-means clustering; a modified FDSK-11 DDS score, which tracks dietary diversity; and the E-NRF score, which tracks the intake of seven key nutrients to encourage in an older population and three to limit. VIP (variable importance in projection) plots of these models showed consistently different features across groups, with higher levels of amino acids and choline-based compounds in lower scores and “unhealthier” clusters much higher in meat and sweet foods.

However, these multivariate models also validated poorly, although the saliva-based OPLS-DA (Orthogonal Projections to Latent Structures Discriminant Analysis) models showed better visual separation and appeared to be less overfit. Together, these results show the benefits of a synergistic approach using multiple instruments towards biomarker discovery, the usefulness of metabolomics-based biomarkers as additional tools for the betterment of nutritional epidemiology studies, and the value of saliva as an under-utilized biofluid in metabolomics.

Future work should focus on the validation of these biomarkers and models in larger and more heterogenous cohorts, with a specific focus on the creation of metabolomics-based multivariate approaches to modelling complex dietary patterns with better cross-validation.