A Comprehensive Evaluation of Taxonomic Classifiers in Marine Vertebrate eDNA Studies

Area of Science:

• Marine biology
• Bioinformatics
• Conservation science

Background:

• Environmental DNA (eDNA) metabarcoding is crucial for marine vertebrate biodiversity surveys.
• Numerous computational tools and bioinformatic approaches exist for eDNA analysis.
• Reliable benchmarks are needed to enhance accuracy and reproducibility in eDNA studies.

Purpose of the Study:

• To comprehensively evaluate nine taxonomic classifiers for marine vertebrate eDNA metabarcoding.
• To assess the performance of three mitochondrial markers (12S rDNA, 16S rDNA, COI) in Australian marine vertebrates.
• To establish best practices for improving the reliability of eDNA-based biodiversity monitoring.

Main Methods:

• Utilized simulated datasets with curated reference and exclusion databases to mimic diverse species compositions.
• Tested nine taxonomic classifiers across 12S rDNA, 16S rDNA, and COI mitochondrial markers.
• Employed positive and negative control datasets to evaluate classifier performance and susceptibility to false positives.

Main Results:

• Identified between 19% and 89% of marine vertebrate species using mitochondrial markers.
• MMSeqs2 and Metabuli outperformed BLAST for 12S and 16S rDNA markers (10-11% higher F1 scores).
• Naive Bayes Classifiers (e.g., Mothur) excelled, except for COI where MMSeqs2 showed superior performance.
• MMSeqs2 and BLAST demonstrated lower susceptibility to false positives compared to Kraken2.

Conclusions:

• Recommends MMSeqs2 for taxonomic classification of marine vertebrates due to improved species assignment and reduced false positives.
• Highlights the importance of classifier selection for accurate eDNA biodiversity assessment.
• Aims to enhance the reliability of eDNA as a tool for marine vertebrate conservation.