Predictive modeling platform for plant protein ingredients in extruded snack foods

Overview

This research initiative develops predictive models that forecast how plant protein ingredients behave during extrusion processing of snack foods. By combining ingredient composition, formulation variables, and extrusion conditions as inputs, the models predict textural and morphological outcomes of finished products. The platform addresses a key challenge in the snack food industry: the resource-intensive, trial-and-error approach typically required to develop appealing extruded products with added plant proteins. Manufacturers and ingredient suppliers can use these models to select, create, or manipulate novel ingredients for protein-enriched snack formulations, reducing development time and cost.

Technical specifications

Modeling approach:

• Inputs include ingredient composition, viscoelastic behavior, and extruder processing parameters
• Outputs include textural and morphological/structural parameters of finished extrudates
• Linear and multi-class logistic regression models predict functional outcomes from experimental data
• Machine learning techniques enable non-linear data fitting for complex relationships
• Molecular dynamics simulations mimic extrusion conditions to compute protein-starch interactions and protein diffusion rates

Experimental capabilities:

• Evaluation of protein functionality in starch-protein mixtures at various stages of heating and cooling
• Analysis of extrudates with varying protein-to-starch ratios, moisture content, and processing temperatures
• Measurement of texture, expansion, color, protein extractability, particle size, and viscosity
• Complementary viscometer instrumentation to evaluate product behavior at critical extrusion steps

Team expertise:

• Established track record in molecular modeling and extrusion research
• Experience studying protein functionality in various mixtures

Technology readiness level

The project is at an early-to-mid research stage. The team has a demonstrated history of conducting related molecular modeling and extrusion work, as evidenced by prior publications. The current proposal seeks to generate foundational datasets by producing extrudates with systematically varied protein-to-starch ratios, moisture content, and processing temperatures, then building and validating predictive models. A novel viscometer acquisition is planned to enable complementary measurements critical to model development. Future validation includes fitting experimental data to regression models, applying machine learning for non-linear fitting, and refining models through molecular dynamics simulations. The approach builds on established basic science modeling methods adapted to food structure and formation applications.