fungi-its-barcodes

Fungal internal transcribed spacer (ITS) reference sequences for 6-level taxonomic classification (phylum → class → order → family → genus → species), in the UNITE convention.

Files

File	Sequences	Purpose
trainset.fasta	~1.5 M	Training references
trainset_valid.fasta	—	Validation split
test1.fasta	4,439	Test 1 — Yeast-enriched
test2.fasta	11,619	Test 2 — Filamentous fungi
test3.fasta	367,337	Test 3 — large-scale benchmark
top50.fasta	2,024	Demo subset

Format

FASTA headers encode the full taxonomic path; unknown ranks are marked ?:

>KY106088|k__Fungi;p__Ascomycota;c__Saccharomycetes;o__Saccharomycetales;f__Saccharomycetaceae;g__Zygotorulaspora;s__Zygotorulaspora_mrakii|SH1312607.09FU

Class counts (trainset)

phylum 18 · class 70 · order 231 · family 791 · genus 3,695 · species 14,742

Source

Reference sequences originate from the UNITE + INSD dataset for Fungi (Abarenkov et al. 2023, PlutoF, doi:10.15156/bio/2938065). The train/validation/test partition and the UNITE-style taxonomic headers follow the MyCoAI data release (Romeijn et al. 2024, Molecular Ecology Resources, doi:10.1111/1755-0998.14006; github.com/MycoAI/MyCoAI).

If you use this dataset, please cite both sources above.

Usage

Fine-tuned model: FungiHelixSeek-Taxonomy — species accuracy 70.8% on the filamentous test set (test2.fasta, cleanly-labelled reference subset).

python scripts/species_classification/predict.py \
    --model_path models/FungiHelixSeek-Taxonomy \
    --input_fasta test2.fasta \
    --output predictions.csv

Training recipe and full metrics: scripts/species_classification on GitHub.

License

CC-BY-NC 4.0 (research use; contact authors for commercial licensing).