Multi-objective Bayesian optimisation on the textural properties of plant-based meat analogues through high-moisture extrusion

Abstract

Plant-based high-moisture meat analogues (HMMA) have gained market traction as sustainable alternatives to conventional meat. However, broader consumer adoption hinges on their ability to closely mimic the textural and sensory properties of real meat. While textural characterisation of HMMA, including hardness, springiness, chewiness, and cutting force, has been reported, optimising HMMA production to achieve these properties remains a challenge. This study presents a human-guided multi-objective Bayesian optimisation (MOBO) framework to optimise the textural properties of HMMAs that were produced by high-moisture extrusion cooking. A Gaussian process surrogate model was employed to map the relationship between extrusion parameters (moisture content, barrel temperature, and screw speed) and HMMA textural properties, and the MOBO framework used this surrogate model to generate promising candidates for the subsequent trials. The objective was for HMMAs to attain the texture of cooked chicken breast meat. Our results demonstrated the effectiveness of MOBO in guiding the optimisation process. The Pareto front, a set comprising optimal trade-off values of hardness and cutting force such that improving one necessarily degrades another, was monitored across multiple trials, converging towards the desired target values for hardness and cutting force with a range of difference between −5.23% and −7.10% and −14.67%–7.33%, respectively. This proof-of-concept framework lays the groundwork for future studies exploring more complex extrusion parameters and expanding the range of targeted meat analogues.