AI-powered foodome characterization

Expected Outcome: Project results are expected to contribute to all the following expected outcomes: an Artificial Intelligence (AI) tool able to accelerate the analysis of complex foodomics data to speed up the identification, structural characterisation, and quantification of chemicals present in food; improving capacity and synergies in foodomics research between academia, research infrastructures and industry. Scope: Through food and beverages, the population is exposed to an immense variety of chemicals (as contaminants and as ingredients). The “foodome” is the complete set of compounds present in a food sample at a given time; it includes both endogenous biomolecules (produced through a species' metabolic processes) and exogenous compounds (natural and synthetic substances from production, processing, handling, and packaging). The current knowledge is largely confined to a narrow set of approximately 150 food components catalogued by traditional databases. These components represent only a very small fraction of the total chemical complexity present in foods, while thousands (>100,000) of compounds and contaminants remain largely unquantified and underexplored. Understanding this “nutrient dark matter” would require identifying, cataloguing and quantifying the full spectrum of substances present in our food and present in accessible, AI-ready databases. The need for such information is today a bottleneck to make the nutrition field predictive and data driven, and has the potential to make food processing more efficient in bringing new foods items on the market. Efforts to create such datasets are underway but owing to the quantity of samples (hundreds of thousands of compounds across thousands of foods), such databases often rely on high-throughput experimental methods which do not allow a precise characterisation of the molecules identified. Proposals should address the following activities: develop and test validated AI algorithms capable of interpreting and annotating high-throughput foodomics data, for instance from untargeted mass-spectrometry or nuclear magnetic resonance; the AI algorithms should be capable to predict or infer the presence and concentration of compounds based on (incomplete) foodomics data via innovative approaches, for instance by considering phylogenic relationships between species. Moreover, this should help identify analytical methods and potential food safety hazards based on their similarity to known compounds; develop a European research and innovation network between Member States and Associated Countries in food …