Microbiome-based classification models for fresh produce safety and quality evaluation.

Area of Science:

• Microbiome analysis
• Bioinformatics
• Food safety and quality

Background:

• Limited statistical power in microbiome studies due to small sample sizes and read loss hinders identification of produce safety (PS) and quality (PQ) indicators.
• Existing amplicon sequence variant (ASV) methods face challenges in differentiating produce phenotypes effectively.

Purpose of the Study:

• To explore a machine learning-based k-mer hash analysis strategy for identifying DNA signatures predictive of PS and PQ.
• To compare the performance of k-mer hash analysis against the ASV strategy for classifying produce safety and quality.

Main Methods:

• Applied machine learning models, specifically random forest (RF), using k-mer hash and ASV data sets.
• Integrated individual and multiple data sets for classification.
• Developed RF-based models using feature-selected ASV-based taxonomic data sets for comparison.

Main Results:

• Random forest classifiers using 7-mer hash data sets showed significantly higher classification accuracy for PS and PQ compared to ASV data sets.
• Integrated k-mer hash strategy improved classification performance but was surpassed by feature-selected ASV taxonomy for PS classification.
• RF feature selection identified 480 PS and 263 PQ indicators.

Conclusions:

• The k-mer hash strategy offers a powerful approach for microbiome data analysis in food safety and quality.
• ASV-based taxonomy remains efficient for PS and PQ classification and identifying key taxa, despite limitations in computational resources.
• This study provides a foundation for comparing microbiome sequencing data sets using machine learning for microbial safety and quality applications.