Development of non-linear prediction model for starch blending

Abstract

Starch blending is a common method for achieving ‘clean label’ with altered properties. However, blending is typically non-additive, making it challenging to predict property changes after mixing. Therefore, this study not only investigated the physicochemical properties of blended starch but also predicted them from individual starches. Six starch samples (rice, corn, potato, tapioca, mungbean, waxy rice) were selected and assessed for individual intrinsic properties (granule size, amylose content, crystallinity, amylopectin chain length). Starch binary blends were then prepared from six starch samples at different blending ratios (100:0, 75:25, 50:50, 25:75, 0:100). Non-additive behaviors were shown in pasting stability, gel storage modulus, and hardness. Notably, the gel storage modulus exceeded expectations, increasing by 140.5% in rice-potato (50:50) and 116.2% in potato-tapioca (50:50) blends. The study also revealed strong correlations among starch properties, suggesting the potential to predict physical properties (e.g., gel hardness, storage modulus) based on intrinsic properties. Developed non-linear models based on the data accurately predicted properties beyond the test set, with a high coefficient of determination (R2 ≒ 0.9). As a result, the models proposed in this study successfully predict physical properties of blended starch, benefiting the food industry interested in clean label starch production and application.