with epidemiological studies and food classification could lead to an automated and practical pipeline capable of systematically improving population diet and individual health. Furthermore, the systematic addition of chemical concentrations for additives and processing byproducts in all foods will enable the construction of anthat is completely unsupervised and independent from any manual classification.
We selected FNDDS 2009–2010 as the main data source for training FoodProX because it allowed us to combine the NOVA labels assigned by Steele et al. inwith one the most comprehensive nutrient panels available for population studies.
The availability of a large nutritional panel in FNDDS 2009–2010 enabled us to train FoodProX using various subsets of nutrients. The widest panel encompasses 99 nutrients, including flavonoid measurements developed for NHANES 2007–2010.