Evaluation of Applied Statistical Models to Optimize Ultrasound-Assisted Extraction of Proteins From Hemp Press Cakes

Abstract

This study aims at evaluating the efficacy of three developed mathematical models to optimize the physical parameters of ultrasound-assisted extraction applied to hemp press cake (HPC) fractions. A discard of industrial hemp seed oil manufacturing process, HPCs were fractioned based on their particle size and subjected to ultrasonic pretreatment with all possible combinations of ultrasonic power, ultrasonic bath temperature, and ultrasonic application time on the samples. The devised models differed in their algorithmic approach, and each resulted in a different degree of fit for the data sets. Soluble protein yield was chosen as the variable to assess the performance of each applied model. Our results indicate that regression models such as Smoothing Splines Regression and Kernel Regression produced highly favorable results for the degree of fitness of data to the models, with R² values of 0.949 and 0.953 for HPC fraction HPC-S, particle size < 500 μm, and 0.897 and 0.903 for the HPC-L fraction, particle size > 500 μm, respectively. These models also predicted higher values of soluble protein yield, reaching 25.39 mg/mL for HPC-S and 17.82 mg/mL for HPC-L, under optimized conditions of ultrasonic power, temperature, and application time. The findings of this study also highlight that such modeling procedures can be applied universally to optimize ultrasound-assisted extraction of plant-origin food materials.