Modeling growth limits of Bacillus spp. spores by using deep-learning algorithm

Area of Science:

• Food Microbiology
• Computational Biology
• Predictive Modeling

Background:

• Bacillus spores pose food safety challenges due to their resistance.
• Predicting microbial growth boundaries is crucial for effective food preservation.
• Existing models may not fully capture complex interactions influencing bacterial growth.

Purpose of the Study:

• To develop and compare predictive models for Bacillus spore growth/no growth boundaries.
• To evaluate the performance of logistic regression, neural networks, and deep learning.
• To assess model applicability in diverse food matrices.

Main Methods:

• Developed growth/no growth boundary models using logistic regression, neural network, and deep learning.
• Tested models against 317 conditions varying pH, water activity, acetic acid, and lactic acid.
• Validated models using independent experimental data from tryptic soy broth and clam soup.

Main Results:

• All models (logistic regression, neural network, deep learning) described Bacillus growth boundaries.
• Neural network and deep learning models outperformed logistic regression in complex scenarios.
• Deep learning demonstrated superior prediction accuracy with lower variability on independent data.

Conclusions:

• Deep learning provides a robust and flexible approach for bacterial growth boundary modeling.
• These models can enhance food safety by enabling precise control of microbial growth.
• The developed deep learning procedure is valuable for safe and adaptable bacterial control strategies.