Health-driven personalized metabolic models of postprandial glucose responses to mixed meals

Anurag Das, Ghady Nasrallah, Sicong Huang, Bobak J Mortazavi, Ricardo Gutierrez-Osuna

Confirmation: I have read and agree with the IEEE BSN 2025 conference submission's policy on behalf of myself and my co-authors.

Keywords: Continuous glucose monitors, diet monitoring, triplet loss, metabolic syndrome

Abstract: The relationship between the macronutrient composition of a meal and the resulting post-prandial glucose response is complex given the large inter-individual differences in metabolism. We present JointCGMacros, a computational model that learns a joint embedding of meal macronutrients and postprandial glucose, mediated by demographics, metabolic health, and gut microbiota variables. The model extracts parallel embeddings from (1) postprandial glucose responses to a meal, and (2) the meal's macronutrient composition conditioned on health parameters using a triplet loss. The macronutrient embedding is an interpretable parametric expression that captures how health parameters modulate the effect of individual macronutrients. We evaluated the model on an experimental dataset containing postprandial glucose responses to a variety of mixed meals from subjects with different metabolic health status (healthy, pre-diabetes, type 2 diabetes). JointCGMacros significantly outperforms a model that attempts to predict macronutrients directly from postprandial glucose. These findings may lead to the development of automatic dietary monitoring using off-the-shelf wearable devices.

Track: 3. Signal processing, machine learning, deep learning, and decision-support algorithms for digital and computational health

NominateReviewer:

Name: Ghady Nasrallah
Email: [e-mail]

Submission Number: 118