An in-depth evaluation of metagenomic classifiers for soil microbiomes

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• Metagenomic projects like human microbiome and TARA Oceans reveal microbial biome complexity.
• Soil microbiomes present unique challenges requiring specialized methods for accurate analysis.
• A key limitation is the lack of soil-specific reference databases for metagenomic classifiers.

Purpose of the Study:

• To evaluate the performance of leading metagenomic classifiers on soil microbiome data.
• To assess the impact of custom, soil-specific databases versus default databases.
• To determine optimal parameters for accurate taxonomic classification of soil microbial communities.

Main Methods:

• Generated a custom in-silico mock community of soil microbial genomes.
• Simulated shotgun sequencing data from this mock community.
• Evaluated Kraken2 (with Bracken and custom/default databases), Kaiju, and MetaPhlAn using default databases.

Main Results:

• Classifiers using databases tailored to specific soil taxa showed fewer errors than broad databases.
• Optimal performance was achieved with a relative abundance threshold of 0.001% or 0.005%.
• Kraken2 with Bracken and a custom GTDB-TK database demonstrated superior precision, sensitivity, and F1 score, classifying 99% of in-silico and 58% of real-world soil reads.

Conclusions:

• In-silico methodological optimization offers significant advantages for soil metagenomics.
• Classifier and database selection critically influence microbial taxonomic profiling.
• Kraken2 with Bracken and a custom GTDB-TK/fungal database at 0.001% abundance threshold provides optimal accuracy for soil shotgun metagenome analysis.