Just keep it simple? Benchmarking the accuracy of taxonomy assignment software in metabarcoding studies.

Abstract

How do you put a name on an unknown piece of DNA? From microbes to mammals, high-throughput metabarcoding studies provide a more objective view of natural communities, overcoming many of the inherent limitations of traditional field surveys and microscopy-based observations (Deiner et al., 2017). Taxonomy assignment is one of the most critical aspects of any metabarcoding study, yet this important bioinformatics task is routinely overlooked. Biodiversity surveys and conservation efforts often depend on formal species inventories: the presence (or absence) of species, and the number of individuals reported across space and time. However, computational workflows applied in eukaryotic metabarcoding studies were originally developed for use with bacterial/archaeal data sets, where microbial researchers rely on one conserved locus (nuclear 16S rRNA) and have access to vast databases with good coverage across most prokaryotic lineages - a situation not mirrored in most multicellular taxa. In this issue of Molecular Ecology Resources, Hleap et al. (2021) carry out an extensive benchmarking exercise focused on taxonomy assignment strategies for eukaryotic metabarcoding studies utilizing the mitochondrial Cytochrome C oxidase I marker gene (COI). They assess the performance and accuracy of software tools representing diverse methodological approaches: from "simple" strategies based on sequence similarity and composition, to model-based phylogenetic and probabilistic classification tools. Contrary to popular assumptions, less complex approaches (BLAST and the QIIME2 feature classifier) consistently outperformed more sophisticated mathematical algorithms and were highly accurate for assigning taxonomy at higher levels (e.g. family). Lower-level assignments at the genus and species level still pose significant challenge for most existing algorithms, and sparse eukaryotic reference databases further limit software performance. This study illuminates current best practices for metabarcoding taxonomy assignments, and underscores the need for community-driven efforts to expand taxonomic and geographic representation in reference DNA barcode databases.