Prediction of zeta potential of amaranth protein isolate via artificial neural networks using the electrostatic surface potential of 11S proglobulin

Abstract

Plant proteins have attracted significant attention due to various health concerns and food safety issues related to animal-based proteins. Different physical and chemical approaches have been applied to plant proteins to improve their functionality, including chemical, physical, or combination of treatments. One of the main properties of interest is the zeta potential as a measure of surface charge. It can be used to optimize the suspension formulations, estimate the emulsion stability, or predict the food surface interactions. Here, we used a single protein database sequence of a plant protein (11S proglobulin) to obtain the net electrostatic surface potentials at different pH. These values were used for modelling a primary neural network to find a correlation between the measured zeta potential and the calculated electrostatic potential values. The network created by this approach had a high correlation coefficient (R = 0.99) for predicting the electrostatic or zeta potential at various pH for the Amarantin. The study demonstrates the potential use of an artificial neural network and its analysis to predict zeta potential values over a pH range, once the network is trained with appropriate datasets, which can potentially be implemented over a range of other plant proteins. The limitation in theoretical protein models, including the complexity of the protein structures and pH-dependent changes of amino acids, must be considered when developing such models. This approach can be explored further to consider protein interactions in the presence of buffer or with added electrolytes, including changes in the surface charge of the molecules.