Predictive Fermentation Control of Lactiplantibacillus plantarum Using Deep Learning Convolutional Neural Networks

Area of Science:

• Biotechnology
• Microbial Fermentation
• Artificial Intelligence in Bioprocessing

Background:

• Microbial fermentation, particularly with Lactiplantibacillus plantarum, is complex due to nonlinear growth dynamics.
• Traditional monitoring methods lack the foresight for early intervention in fermentation processes.
• Predicting fermentation outcomes requires advanced analytical approaches beyond conventional monitoring.

Purpose of the Study:

• To develop a deep learning-based system for classifying fermentation trajectories and predicting final cell counts.
• To utilize early-stage fermentation data (first 24 hours) for predictive modeling.
• To enhance real-time control and efficiency in industrial probiotic production.

Main Methods:

• Convolutional Neural Networks (CNNs) were employed for classification of fermentation outcomes.
• Real-time fermentation parameters (pH, temperature, dissolved oxygen) were recorded and transformed into time-series features.
• A dataset of 52 fermentation runs was used for training and validation of the predictive model.

Main Results:

• The CNN model achieved a high classification accuracy of 97.87% for predicting fermentation success, semi-success, or failure.
• The deep learning model outperformed benchmark models like LSTM and XGBoost in predictive accuracy.
• Early predictions from the model facilitated manual interventions, preventing batch failures and improving suboptimal fermentations.

Conclusions:

• Deep learning offers a robust and scalable framework for real-time fermentation control.
• The proposed CNN-based system demonstrates significant potential for enhancing efficiency and reducing costs in industrial probiotic production.
• Predictive modeling using early fermentation data is crucial for optimizing bioprocess outcomes.