P571 - METABOLIC AVATAR WITH HYBRID MODELING: BEYOND AI FOR PERSONALIZED CLINICAL NUTRITION

P571

METABOLIC AVATAR WITH HYBRID MODELING: BEYOND AI FOR PERSONALIZED CLINICAL NUTRITION

C. Serantoni¹, M. M. De Giulio^1,2, T. Marchetti¹, A. Riente², D. Hatem², G. Maulucci^1,2,*

¹Physics, Fondazione Policlinico Gemelli IRCCS, ²Physics, Università Cattolica del Sacro Cuore, Rome, Italy

Rationale: While artificial intelligence has opened new frontiers in personalized nutrition, purely data-driven models often lack physiological interpretability[1,2,3]. We present a metabolic avatar based on hybrid modeling—combining AI with compartmental and network-based representations of human metabolism—to move beyond prediction and toward understanding. We present a personalized metabolic avatar that integrates data-driven learning with compartmental modeling and curated metabolic networks, enabling both accurate predictions and mechanistic explanations of individual nutritional responses.

Methods: The system employs a multi-compartment framework to simulate nutrient flows across organs and tissues, coupled with stoichiometrically constrained models of key metabolic pathways (glycolysis, β-oxidation, TCA cycle). Inputs include dietary macronutrient profiles, energy expenditure, and anthropometric variables. The hybrid model outputs both weight evolution and underlying metabolic fluxes.

Results: In a 100-day longitudinal study on 30 adults (60% female), the avatar achieved a mean RMSE of 0.25 ± 0.12 kg for daily weight prediction,  resting heart rate and other physiological variables, with simulation runtimes below 15 seconds. The compartmental structure enabled mapping of nutrient utilization across physiological domains. The integration of metabolic modeling with AI facilitated both individual adaptation and biological transparency, surpassing black-box limitations.

Conclusion: This hybrid metabolic avatar represents a new generation of personalized nutrition tools, combining explainability with precision. By going beyond AI, it supports clinical decisions in obesity management, chronic disease nutrition, and metabolic optimization.

References: [1] A. Abeltino;...; G. Maulucci , Computers in Biology and Medicine, 2025-04

[2] A. Abeltino; ... G. Maulucci, Nutrients, 2023-02-27

[3]A Abeltino; ...; G Maulucci, Nutrients, 2022-08-26 

Disclosure of Interest: None declared