Machine Learning-Enabled Rapid Assessment of Plant-Based Protein Digestibility Through Physicochemical Profiles

Area of Science:

• Food Science
• Biotechnology
• Sustainable Agriculture

Background:

• Plant-based proteins are sustainable but face low digestibility challenges.
• Current methods for assessing protein digestibility are time-consuming and require large samples.
• These limitations hinder efficient optimization of plant protein sources.

Purpose of the Study:

• To develop a rapid, low-sample method for predicting plant protein digestibility.
• To identify key physicochemical properties influencing protein digestibility.
• To accelerate the screening and development of improved plant-based proteins.

Main Methods:

• Systematic characterization of 23 plant protein isolates across multiple physicochemical dimensions.
• Training a feedforward neural network using augmented data for digestibility prediction.
• Developing a streamlined three-feature model based on identified key indicators.

Main Results:

• The deep learning model achieved high prediction accuracy (R² = 0.91) on independent datasets.
• Identified alpha-helix content, random coil content, and solubility as critical digestibility indicators.
• The streamlined model reduced assessment time by 80% and sample requirement by 99%.

Conclusions:

• A transformative framework for rapid plant protein digestibility assessment has been established.
• This method accelerates the screening of novel plant protein sources and modification strategies.
• The findings support enhanced nutritional bioavailability and sustainable food system transitions.