Taxonova

Taxonova assigns 16S rRNA and rRNA operon amplicon sequences to a lineage from domain to species, with a confidence at each rank, and stops at the deepest rank it is confident about.

Classify your sequences

No installation: upload a FASTA file to the web app.

Command line (Python 3.10 or later):

pip install taxonova
taxonova classify reads.fasta --model v4

The model is downloaded on first use (about 250 MB, once). Results are written to reads.taxonova.tsv.

Models

The 16S models are trained on Greengenes2, and species names follow GTDB in all models. The V4-V5 model has known limitations; see models. Run taxonova models to see which models are downloaded.

Output

One row per sequence: a label and a confidence for each rank (columns for phylum to family not shown here).

sequence_id   domain    domain_confidence  ...  genus       genus_confidence  species                 species_confidence
f1eb3726...   Bacteria  1.0000             ...  Prevotella  0.9988            Prevotella pectinovora  0.9988

A rank below its confidence threshold, and every rank below it, reads Unclassified. Each model sets its own thresholds unless you give one.

Options you may want

taxonova classify reads.fasta -m v3v4 -o results.tsv   # model and output file
taxonova classify reads.fasta -m v4 --confidence 0.9   # set the species threshold yourself
taxonova classify reads.fasta -m v4 --threads 8        # more CPU threads
taxonova classify reads.fasta -m v4 --genus-threshold 0.6   # set one rank

--confidence sets the species threshold, and the other ranks scale with it; without it, the model's calibrated thresholds are used (output and thresholds). From Python: taxonova.classify("reads.fasta", model="v4") returns a pandas DataFrame, with the same thresholds as the command line.

Links

• Documentation: models, output and thresholds, Python, training your own model
• Changelog
• Paper and the benchmark used in it, TaxoBench
• Report a problem or ask for a model

Citation

If you use Taxonova, please cite:

Kakabra K, Patel S, Temko A, Claesson MJ. Taxonova: deep learning enables fast and confidence-aware microbiome species-level amplicon classification. PLoS Comput Biol. 2026;22(10):e1014912. doi:10.1371/journal.pcbi.1014912

Taxonova is developed in the Claesson Lab, University College Cork and APC Microbiome Ireland, and released under the MIT licence.