Prediction of reduction behavior by heating and strain variability of Campylobacter jejuni using amino acid...

Predicting bacterial behavior and strain variability is essential for quantitative microbial risk assessments in food safety. The growing availability of whole-genome sequencing (WGS) data enables deeper insights into microbial thermotolerance. However, capturing the reduction behavior of Campylobacter jejuni during thermal inactivation, while accounting for strain variability, remains challenging. We aimed to develop a machine-learning model leveraging amino acid phylogenetic information from WGS data to predict the reduction behavior and strain variability of C. jejuni during thermal inactivation. We developed a machine learning model utilizing 38 complete genomes of C. jejuni and their parameters of modified Weibull models describing the reduction behaviors heated at 55 °C. Correlation analyses identified genes that could be relevant to thermotolerance and highlighted genes potentially linked to thermotolerance. Leave-one-out cross-validation yielded a root mean square error of 0.83 log. Strain variability was further estimated using 679 genomes from genomic databases. Strain variability exhibited a multimodal distribution with one prominent peak and three minor peaks, indicating that traditional unimodal distributions could not fully represent the variability in C. jejuni thermal reduction. The machine-learning model effectively predicted reduction behavior and strain variability of C. jejuni during thermal inactivation using WGS data. Nonetheless, its prediction range is limited by the diversity of the training set, suggesting that broader genomic datasets could enhance accuracy. These findings provide a pathway for improved microbial risk assessments through genomic data integration.