Characterising the suitability and limitations of metagenomic tools for the detection and discovery of plant viruses

Abstract

Detecting the presence of viral genomes within a plant tissue sample is a vital task, with implications for food security, ecological networks, and biotechnological discovery. Having confidence that a negative is a true negative is necessary, but also difficult - could a novel virus or viroid pass undetected? In the last decade, metagenomic approaches have become widely used for this task, with many diverse methodologies being proposed and becoming implemented. Each is known to have advantages and disadvantages, but the exact boundaries of their limits of detection, and the factors that influence them, have only been studied within restricted subsets of tools. Successful metagenomic detection of viruses and viroids face three main barriers - the divergence of viral genomes, the presence of low-titre viral genomes, and sparsity of taxonomy within our viral reference databases. In this thesis we develop a benchmarking methodology that allows us to compare a diverse set of approaches in metagenomic viral detection and discovery, and determine the factors that influence their limitations. For this end, we inititally create a novel program for the determination of pairwise substitution distance between short, highly divergent genomes. This program, Mottle, is able to successfully quantify substitution distances further than current alternatives. We then use this as a metric, along with other controlled parameters, to determine which software are able to overcome which barriers. We find that there is a trade-off between performance at high divergence and low read depth, with reference sparsity acting as a modulator. Crucially, no approach showed success at detection when all barriers were high. Finally, we apply these tools to previously seen and novel metagenomic datasets, to compare their outputs, and to synthesise conclusions informed by multiple approaches.