Assessment of current taxonomic assignment strategies for metabarcoding eukaryotes

Area of Science:

• Biodiversity science
• Molecular ecology
• Bioinformatics

Background:

• Metabarcoding is vital for biodiversity surveys and biomonitoring.
• Accurate taxonomic assignment of DNA sequences is a significant analytical challenge.
• Selecting appropriate methods is critical for reliable species composition data.

Purpose of the Study:

• To evaluate the performance and accuracy of seven taxonomic assignment methods.
• To assess the impact of parameter optimization on assignment quality.
• To compare different approaches using mock communities with varying complexity.

Main Methods:

• Utilized mock communities with controlled species and specimen abundance.
• Employed cytochrome c oxidase subunit I (COI) barcode fragments for analysis.
• Benchmarked seven distinct taxonomic assignment methods, including machine learning and sequence similarity approaches.

Main Results:

• BLAST top hit, when optimized, performed comparably to more complex methods.
• Machine learning methods demonstrated higher sensitivity to reference database completeness and heterogeneity.
• Assignment accuracy was influenced by species and specimen counts (query compositional heterogeneity).

Conclusions:

• Realistic mock communities are essential for optimizing taxonomic assignment parameters.
• Careful selection and curation of comprehensive reference databases are crucial.
• QIIME, BLAST, or Lowest Common Ancestor (LCA) methods, with parameter tuning, are recommended for diverse communities, especially in data-limited regions.