382: Characterization and Tailoring of Faba Bean Based Extruded Plant-Based Meats Using Ingredient Functionality and Machine Learning Analyses

Introduction

To improve consumer acceptance of plant-based meat research is needed on unique formulations and process optimization to achieve desired textures. This study investigated optimum extrusion process moisture for faba bean concentrate (FBC) based formulations, in combination with pea protein isolate (PPI), soy protein isolate (SPI), wheat gluten and soy protein concentrate (SPC). Protein physicochemical characterization and machine learning techniques were employed to bridge formulation properties with product microstructure and quality attributes.

Methods

Three formulations (45% FBC, 44% PPI/SPI/Gluten and 11% SPC) were processed using a pilot-scale twin-screw extruder with varying in-barrel moisture (IBM). Specific mechanical energy (SME), bulk density (BD), water holding capacity (WHC) and microstructural properties were evaluated. Proteins were evaluated using rapid viscoanalyzer (RVA) and other tests for hydrophilic and hydrophobic interactions including water absorption capacity and least gelation concentration. Microstructure images obtained using a Nikon-D750 camera were segmented and analyzed using the Roboflow platform, employing Region-based Convolutional Neural Networks (R-CNN) and You Only Look Once (YOLO) algorithms. Image augmentation enhanced prediction reliability, yielding fibrousness scores on a 10-point scale.

Results

SME (596-920 kJ/kg) decreased with increasing IBM, leading to higher BD and reduced porosity. Use of Cold-swelling proteins (SPI) in combination with FBC and with higher SME (795-918 kJ/kg) produced less fibrous structures characterized by higher porosity (bulk density 151-210 g/L) and water holding capacity (WHC) (2.61-4.85 g/g). In contrast, other cold-swelling proteins (PPI) with lower SME (595-751 kJ/kg) and non-cold-swelling proteins (Gluten) resulted in layered, fibrous products that exhibited lower porosity (bulk density 238-333 g/L) and WHC (1.26-2.72 g/g). Machine learning was used to effectively quantify microstructural features such as fibrousness, which was found to inversely correlate with porosity. Data interpretation allowed enhanced understanding of relationships between protein functionality, process and plant-based product quality.

Significance

This research advances the ability to design plant-based meat with diverse microstructures and textures using a range of proteins and extrusion conditions. Machine learning based microstructural analysis offered novel insights into formulation-structure relationships, paving the way for improved product development and consumer acceptance of sustainable alternatives to animal proteins.

Authors: Shirin Sheikhizadeh, Abdullah Aljishi, Sanjoy Das, Yonghui Li, Sajid Alavi