Comparative analysis of metagenomic classifiers for long-read sequencing datasets

Area of Science:

• Metagenomics
• Bioinformatics
• Computational Biology

Background:

• Long-read sequencing is increasingly vital for metagenomics.
• Comprehensive assessment of metagenomics classification tools at the species level is needed.

Purpose of the Study:

• To benchmark various metagenomics classification tools, including kmer-based, mapping-based, and general-purpose mappers.
• To evaluate tool performance across diverse datasets, including synthetic, mock, and real gut microbiomes, under various conditions.

Main Methods:

• Evaluated over 20 pipelines using nucleotide and protein databases.
• Tested 13 selected pipelines on seven synthetic datasets and three mock community datasets.
• Included six real gut microbiome datasets in the benchmark.

Main Results:

• General-purpose mappers (Minimap2, Ram) showed comparable or better accuracy than classification tools, but were slower.
• Kmer-based tools were faster but less accurate; protein databases underperformed nucleotide databases.
• Tool performance varied with read length, database completeness, host DNA presence, and presence of unknown or related species.

Conclusions:

• Kmer-based tools are suitable for fast long-read metagenomics analysis; mappers offer higher accuracy at a slower speed.
• Analyzing complex samples requires a combination of different tool types and databases.
• Continuous improvement of tools and databases is crucial, especially for handling host DNA and novel species.