Computational Nutrition 2.0: From Static Models to Closed-loop Agentic Systems for Personalized Nutrition.

Abstract

Computational Nutrition is an emerging interdisciplinary field that addresses complex challenges in nutrition and health through computational methods and multimodal data. Major research directions include 1) prediction of personalized metabolic responses to foods, 2) causal inference and individualized treatment effects in nutrition and diseases, 3) dynamic assessment and monitoring of diet-related disease risks, and 4) simulation and evaluation of nutrition policies, collectively driving a paradigm shift in traditional nutrition. This paper introduces Computational Nutrition 2.0, showing how artificial intelligence (AI) agents can advance these directions. Most existing models are static, passive, and unidirectional. AI agents overcome these limitations by continuously learning from new data, autonomously perceiving physiological and behavioral states, and engaging in ongoing dialogue with users. In personalized metabolic response prediction, agentic digital twins offer a potential pathway toward dynamic dietary intervention, though this vision remains aspirational and requires further advances in agent technology, sensor reliability, and regulatory approval. In causal inference, nutritional causal agents enable in silico clinical trial simulation through automated causal graph construction and counterfactual reasoning, addressing the static and passive limitations of traditional approaches. In disease risk monitoring, agent-driven perception systems leverage non-invasive sensors for continuous monitoring and early warning, overcoming the passivity of conventional risk assessment. In nutrition policy, multi-agent systems generate Pareto-optimal policy portfolios, revealing trade-offs across health, environmental, economic, and equity dimensions. Computational Nutrition 2.0 advances the field from static models to closed-loop agentic systems for personalized nutrition, representing a paradigm shift from prediction to dynamic, proactive, and adaptive intervention. The era of AI agents offers nutrition an unprecedented opportunity to move from informing to partnering.