Encoder-decoder neural networks for predicting future FTIR spectra - application to enzymatic protein hydrolysis

Area of Science:

• Food Science and Technology
• Biotechnology
• Process Engineering

Background:

• Optimizing the conversion of food-processing by-products into value-added ingredients requires precise control over raw materials, enzymes, and process conditions.
• Variability in raw material batches and lack of characterization hinder consistent yield and economic viability.
• Online or at-line monitoring of enzymatic reactions is crucial for managing batch variations and ensuring process efficiency.

Purpose of the Study:

• To investigate the application of deep neural networks (DNNs) for predicting the future state of enzymatic hydrolysis.
• To assess the utility of Fourier-transform infrared (FTIR) spectra for monitoring enzymatic reactions in real-time.
• To develop a flexible and transparent tool for process monitoring and control in the bioconversion of food by-products.

Main Methods:

• Utilizing deep neural networks to analyze Fourier-transform infrared (FTIR) spectra of hydrolysates.
• Developing predictive models for the state of enzymatic hydrolysis based on spectral data.
• Integrating predictions of average molecular weight with spectral analysis for comprehensive process insight.

Main Results:

• Demonstrated the potential of DNNs to accurately predict the progression of enzymatic hydrolysis from FTIR spectra.
• Showcased the capability to estimate average molecular weight alongside spectral predictions.
• Established a foundation for a data-driven approach to monitor and control bioprocesses.

Conclusions:

• Deep neural networks, combined with FTIR spectroscopy, offer a powerful method for monitoring enzymatic hydrolysis.
• This approach provides a flexible and transparent tool for proactive process control and optimization.
• Enhances the economic viability of converting food-processing by-products into valuable ingredients by managing batch variations effectively.