71-LB: Personal Metabolic Responses to Food Predicted Using Multi-omics Machine Learning: The PREDICT I Study

Abstract

Background: Postprandial glycemic variations impact diabetes risk and are influenced by food consumed, a focus of major nutritional guidelines. However, a person’s biological characteristics and environment likely influence postprandial response. Methods: In ∼1000 healthy UK (unrelated + DZ/MZ twins) and 100 unrelated U.S. adults, glycemic responses to a variety of meals were tested in clinic and at home (using CGM). Baseline factors included metabolomics, genomics, gut metagenomics and body composition. Genetic contributions to postprandial responses were determined by classical twin methods. Results: Within-individual concordance in metabolic responses (2hr glucose iAUC) for pairs of isocaloric meals at home was moderate to high: high fat (fat = 40g/71% energy) (ICC=0.46, 95% CI:0.37,0.55), high protein (protein = 41g/32% energy) (ICC=0.54, 95% CI:0.44,0.62), high carb (carbohydrate = 95g/76% energy) (ICC=0.65, 95% CI:0.59,0.70), average lunch (carbohydrate = 68g/54% energy) (ICC=0.61, 95% CI:0.53,0.67), and OGTT (carbohydrate = 75g) (ICC=0.66, 95% CI:0.61,0.71). Inter-individual variability for the same meals was (IQRs: 20, 29, 45, 68, 76 mmol/L*min, respectively). Using the twin data, the unadjusted genetic contribution of postprandial glycemic responses in the clinic was 53%. An interim machine learning algorithm predicted 46% of the variation in glycemic responses based on meal content, meal context and the participant’s baseline characteristics, excluding genetic and microbiome features. Conclusions: Variation in metabolic responses to the same meals between healthy people is high, helping explain why ‘one-size-fits-all’ nutritional guidelines often fail. The genetic component to these responses is moderate, which leaves much of the variation potentially modifiable. By collecting information on glucose responses to over 100,000 meals, we now have excellent power to use machine learning to predict which foods optimize responses. Disclosure T.D. Spector: None. S. Berry: Consultant; Self; Zoe Ltd. A.T. Chan: Consultant; Self; Bayer AG, Janssen Pharmaceuticals, Inc., Pfizer Inc. R.J. Davies: Employee; Self; Zoe Global Limited. D.A. Drew: Consultant; Spouse/Partner; PathAI. A.M. Valdes: Consultant; Self; Zoe Ltd Global. Research Support; Self; Pfizer Inc. P.W. Franks: Board Member; Self; Zoe Ltd. Research Support; Self; AstraZeneca, Boehringer Ingelheim International GmbH, Lilly Diabetes, Novo Nordisk A/S, Novo Nordisk Foundation, Sanofi, Servier. Funding UK National Institute for Health Research; UK Wellcome Trust; Zoe Global Limited