Reference-Free Plant Disease Detection Using Machine Learning and Long-Read Metagenomic Sequencing.

Area of Science:

• Plant pathology
• Genomics
• Bioinformatics
• Machine learning

Background:

• Early detection of plant diseases is vital for food security, especially with emerging pathogens exacerbated by climate change.
• Current genomic methods rely on known pathogens, limiting their use for novel or unknown diseases.
• Field diagnostics face challenges with limited database access.

Purpose of the Study:

• To explore reference-free detection of plant pathogens using metagenomic sequencing and machine learning (ML).
• To develop a method for identifying plant diseases without prior knowledge of pathogen genomes.
• To assess the feasibility of ML models for real-time, field-deployable plant disease surveillance.

Main Methods:

• Utilized long-read metagenomes from healthy and infected plants.
• Constructed k-mer frequency tables to train and test eight different ML models.
• Evaluated model performance based on accuracy in classifying individual sequencing reads.

Main Results:

• Random Forest (RF) model demonstrated high accuracy (over 0.90) and short run-time.
• The RF model successfully classified reads from different tomato and grapevine infections.
• A single ML model trained on one pathogen-host system could detect different pathogens on other hosts.

Conclusions:

• Machine learning enables accurate, reference-free detection of plant diseases from metagenomic data.
• This approach holds potential for a universal plant disease surveillance pipeline, adaptable to new and emerging pathogens.
• Further research is needed to address the challenges in applying ML to metagenomics for plant disease diagnostics.