An AI-informed framework for flavor design: How κ-carrageenan sculpts low-sodium myofibrillar protein architecture for aldehyde capture and bitterness masking.

Designing healthy, palatable foods is a grand challenge in materials science, particularly in quest for low-sodium products where flavor is often compromised. The central problem is not merely replacing NaCl, but re-engineering the food matrix itself to manage off-flavors and retain desirable aromas. Here, an AI-informed framework demonstrates how the biopolymer κ-Carrageenan (KC) can sculpt the myofibrillar protein (MP) architecture for enhanced aldehyde capture and bitterness masking in low-sodium systems. Results showed KC acts as a molecular scaffold, transforming the protein from a rigid state into a porous, aggregated network. This remodeling significantly boosted binding for key flavor aldehydes (3-methylbutanal < pentanal < hexanal < heptanal). KC modulated MP-heptanal interactions primarily involved hydrogen bonding, whereas other aldehydes interacted mainly via hydrophobic forces. Concurrently, KC's sulfate groups sequestered K+, masking bitterness while enhancing saltiness and umami. To decipher this interplay, an interpretable machine learning model (R2 > 0.9) quantitatively linked conformational changes to flavor. SHAP analysis then revealed aldehyde concentration, protein solubility, and thiol group availability as the most critical features governing aldehyde absorption and bitterness perception. This work pioneers a structure-driven, Al-based strategy for flavor design, offering a new paradigm for the rational formulation of next-generation healthy foods.