Practical application of self-organizing maps to interrelate biodiversity and functional data in NGS-based metagenomics

Next-generation sequencing (NGS) technologies have enabled the application of broad-scale sequencing in microbial biodiversity and metagenome studies. Biodiversity is usually targeted by classifying 16S ribosomal RNA genes, while metagenomic approaches target metabolic genes. However, both approaches remain isolated, as long as the taxonomic and functional information cannot be interrelated. Techniques like self-organizing maps (SOMs) have been applied to cluster metagenomes into taxon-specific bins in order to link biodiversity with functions, but have not been applied to broad-scale NGS-based metagenomics yet. Here, we provide a novel implementation, demonstrate its potential and practicability, and provide a web-based service for public usage. Evaluation with published data sets mimicking varyingly complex habitats resulted into classification specificities and sensitivities of close to 100% to above 90% from phylum to genus level for assemblies exceeding 8 kb for low and medium complexity data. When applied to five real-world metagenomes of medium complexity from direct pyrosequencing of marine subsurface waters, classifications of assemblies above 2.5 kb were in good agreement with fluorescence in situ hybridizations, indicating that biodiversity was mostly retained within the metagenomes, and confirming high classification specificities. This was validated by two protein-based classifications (PBCs) methods. SOMs were able to retrieve the relevant taxa down to the genus level, while surpassing PBCs in resolution. In order to make the approach accessible to a broad audience, we implemented a feature-rich web-based SOM application named TaxSOM, which is freely available at http://www.megx.net/toolbox/taxsom. TaxSOM can classify reads or assemblies exceeding 2.5 kb with high accuracy and thus assists in linking biodiversity and functions in metagenome studies, which is a precondition to study microbial ecology in a holistic fashion.