3D printing of wheat starch gels for dysphagia diet: Molecular interaction mechanisms and machine learning-assisted prediction of IDDSI levels.

This study investigated how amylose content and intermolecular interactions regulate the 3D printability, structural properties, water distribution, and swallowing suitability of wheat starch gels for dysphagia patients. Gels containing 26-77% amylose were prepared, and hydrogen bonding or hydrophobic interactions were selectively changed using urea or NaCl. Printing tests showed that 43% amylose achieved the best filament continuity and dimensional fidelity (96.18 ± 0.32%), whereas low amylose caused extrusion breakage and high amylose (≥60%) led to structural collapse with printing accuracy decreasing to 90.88-70.36%. Using 1 M urea moderately adjusts the hydrogen bond network, improving extrusion smoothness while maintaining the stability of the printed structure with a printing accuracy of 97.06 ± 0.10%. In contrast, 3 M urea disrupts the network more extensively, increases free water, and reduces printing accuracy. NaCl modifies the hydration environment through the Hofmeister effect, promotes hydrophobic association, and strengthens hydrogen bonds, thereby enhancing the printing stability of the gel. Texture analysis and IDDSI classification further supported these findings: the 43% amylose gel exhibited moderate hardness (89.58 N) and adhesiveness (222.8 N·s), corresponding to IDDSI Level 5 (Minced & Moist), while strong hydrogen bond disruption or excessive amylose produced softer gels classified as Level 4 (Pureed). To comprehensively investigate the complex interrelationships among multiple factors and to establish a reliable performance prediction model, machine learning approaches (LSSVM, PLSR, RF, and BP-ANN) were employed to predict swallowing level based on seven physicochemical variables selected from correlation analysis, which indicated that BP-ANN was the most accurate method for predicting swallowing level with an accuracy of 65.7%. This mechanistic understanding can be extended to food permitted hydrogen bonding modulators, such as polyhydric alcohols, and provides a theoretical foundation for the formulation of 3D printed texture modified foods designed for individuals with dysphagia.