Quantitative and Qualitative Control of Wheat Flour Dough's Mechanical Properties via Dietary Protein Addition

ABSTRACT

Dietary protein fortification enhances bakery nutrition, but its impact on dough mechanics remains unclear. To address this, 432 formulations were designed using eight proteins (up to 70.6% substitution), seven lipid configurations, and salt, with water adjusted (38%–159% baker's percentage). Texture profile analysis (10 mm compression distance) and dynamic rheometry (0.1–80 Hz at 0.5% strain) yielded 19 continuous and tertile‐discretized mechanical target variables. Parallel regression and classification workflows were developed to predict continuous values and tertile‐defined categories (low, medium, and high). Models included Ridge regression for continuous prediction, logistic regression for classification, and Random Forest and Gradient Boosting for both tasks, with performance evaluated using five‐fold cross‐validation. Both paradigms generally predicted rheological properties with higher accuracy ( R ² up to 0.87, classification accuracy up to 89.3%) than textural attributes ( R ² up to 0.78, accuracy up to 82.9%). Permutation importance revealed that soy protein, a pea‐quinoa blend, and bread flour were leading predictors, whereas lipids and salt showed low average predictive weight. Rheological regression models required eight to 10 predictors to retain ≥ 90% of full‐model R ² , whereas hardness, gumminess, chewiness, and resilience required four. Casein's normalized predictive importance was 42.1% higher in classification than in regression, indicating task‐dependent relevance. Rather than demonstrating biophysical causation, this research establishes an internally validated predictive framework. These findings advance formulation science by offering a mathematical basis to quantitatively screen and rationally design composite dough formulations. This predictive tool may support high‐plant‐protein food development by prioritizing formulations for experimental testing within the investigated domain.