Over the past year, AIFS has sponsored a project that has created FoodAtlas, an automated, curated knowledgebase of food chemical composition, using NLP entailment models on approximately 2,000 food-related papers, and further quality control from the team. FoodAtlas currently covers 281 foods and 1268 chemicals. This year, the FoodAtlas team will expand the automated curation AI methods and enrich FoodAtlas to additional entity types, such as food contaminants, diseases, and the corresponding relationships.

In addition, the proposed approach focused on molecular classes will allow high quality expert driven data curation to ensure predictors are being trained on high quality datasets. FoodAtlas will be open to the community to use through APIs, and eventually a user interface with graphics will allow access through a web portal. FoodAtlas and flavor prediction will accelerate research and commercialization of different applications in food on a pre-compete level. Terpene Database will be a novel resource that fills in a gap in literature for a class of molecules that are important in flavor, health and foods. Finally, the team will create an educational module on bioactive molecules in food.