Alignment-free microbiome-based classification of fresh produce safety and quality

Abstract

Background: While next generation sequencing has enriched our knowledge about native microbial populations present in fresh produce, the loss of up to 50-70% of data during the alignment and denoising steps of data preprocessing may lead to the missing of important bacterial species information and decrease our ability to identify species associated with poor produce quality and contamination. Microbial studies are also often limited by small sample sizes, making generalization of results beyond individual studies difficult. Results: In this study, we explored separate strategies to mitigate the challenges of data preprocessing and small sample sizes. First, we explored an alignment-free analysis strategy using k-mer hashes to identify DNA signatures predictive of produce safety (contaminated vs. non-contaminated) and produce quality (good-quality vs. decreasing-quality), and compared it against the amplicon sequence variant (ASV) strategy that uses a typical alignment and denoising step. Random forests (RF)-based classifiers were trained on publicly available fresh produce microbiome datasets with data preprocessed using either the k-mer hash or ASV approach. RF-based classifiers for fresh produce safety and quality using 7-mer hash datasets had significantly higher classification accuracy than those using the ASV datasets, supporting the hypothesis that data preprocessing strategies that keep more data (k-mer hash) retain more useful information about bacterial species than approaches that lose data during preprocessing (ASV). We also demonstrated that integrating multiple datasets together also led to higher classification accuracy compared to those trained with individual datasets. Integrated datasets also enabled the identification of more consistent and generalizable biomarkers (ASV, 7-mer hash, or bacterial taxa) associated with fresh produce safety and quality. Conclusions:The proposed combination of integrating multiple datasets and leveraging an alignment-free 7-mer hash strategy substantially mitigates the loss of sequencing data due to the ASV denoising step and leads to better classification performance for fresh produce safety and quality. Results generated from this study lay the foundation for future studies that wish and need to incorporate and/or compare different microbiome sequencing datasets (generated from different studies or different laboratories) for the application of machine learning in the area of microbial safety and quality of food.