A comprehensive evaluation of taxonomic classifiers in marine vertebrate eDNA studies

Abstract

Environmental DNA (eDNA) metabarcoding is a widely used tool for surveying marine vertebrate biodiversity. To this end, many computational tools have been released and a plethora of bioinformatic approaches are used for eDNA-based community composition analysis. Simulation studies and careful evaluation of taxonomic classifiers are essential to establish reliable benchmarks to improve accuracy and reproducibility of eDNA-based findings. Here we present a comprehensive evaluation of nine taxonomic classifiers exploring three widely used mitochondrial markers (12S rDNA, 16S rDNA, and COI) in Australian marine vertebrates. Curated reference databases and exclusion database tests were used to simulate diverse species compositions, including three positive control and two negative control datasets. Using these simulated datasets, we were able to identify between 19% to 85% of marine vertebrate species using mitochondrial markers. We show that MMSeqs2 and Metabuli generally outperform BLAST with 10% and 11% higher F1 scores for 12S and 16S rDNA markers, respectively, and that Naive Bayes Classifiers such as Mothur outperform sequence-based classifiers except MMSeqs2 for COI markers by 11%. Database exclusion tests reveal that MMSeqs2 and BLAST are less susceptible to false positives compared to Kraken2 with default parameters. Based on these findings, we recommend that MMSeqs2 is used for taxonomic classification of marine vertebrates given its ability to improve species-level assignments while reducing the number of false positives. Our work contributes to the establishment of best practices in eDNA-based biodiversity analysis to ultimately increase the reliability of this monitoring tool in the context of marine vertebrate conservation.

Enhancing biodiversity monitoring efficiency through CRISPR-driven depletion and enrichment of aquatic environmental DNA

“…Likewise, physico-chemical properties offer valuable real-time insights into ecosystem health, but are not capable of measuring organism presence or abundance (Smale et al, 2011). Molecular-based monitoring techniques using environmental DNA (eDNA) for biodiversity detection have grown rapidly over the past decade (Bayer et al, 2024;Blackman et al, 2024;De Brauwer et al, 2023;Ficetola et al, 2008;Garlapati et al, 2019;Taberlet et al, 2018;Takahashi et al, 2023). Environmental DNA biomonitoring has recently been applied to health indices (Borja, 2018;Pawlowski et al, 2022;Wilkinson et al, 2024), as the increased reliability and accuracy of eDNA biomonitoring through continuous methodological innovation has been shown to be comparable to, and even surpassing, previous monitoring records with regards to efficiency and species detection (Baetscher et al, 2023;Elbrecht et al, 2017;Shea et al, 2023;Wilkinson et al, 2024).…”

“…Primer set and bioinformatic pipeline choice can further influence the accuracy of metabarcoding results, with the same primer set behaving differently and not always predictably based on the complexity of an eDNA sample and what species are present, and with different bioinformatic pipelines each having different criteria the data is tested against, thus causing variable results from the same data (Baetscher et al, 2023;Bayer et al, 2024;De Brauwer et al, 2023;McCauley et al, 2024;Pearce et al, 2023;Shea et al, 2023). While metabarcoding holds promise for rapid ecosystem monitoring, reliance on PCR amplification is resulting in false-negative species detections challenges in accuracy and limiting quantitative measurements of species abundance and biomass, thereby presenting challenges for widespread adoption as a monitoring method for management and the knowledge-transfer to end-users (Elbrecht et al, 2017;Ruppert et al, 2019).…”

phyloBARCODER: A Web Tool for Phylogenetic Classification of Eukaryote Metabarcodes Using Custom Reference Databases

“…In metabarcoding, although composition-based methods such as Bayesian classifier ( Wang et al 2007 ) are widely used with stand-alone programs ( Bayer et al 2024 ), a common method of species identification is to quantify similarity between reads and reference sequences using web tools such as NCBI BLAST searches ( Altschul et al 1997 ; Collins et al 2021 ). However, there are several problems ( Fig.…”