BouillaBERT dataset

This archive contains all data necessary to train, validate and test the models described in: Transductive learning from a transformer foundation model improves the taxonomic classification of eDNA metabarcodes The datasets include training, validation, and test partitions for four eDNA markers commonly used in fish metabarcoding: Teleo, MiFish-U, Berry, and Ac16S.For each marker, sequences are split into six folds. The split configuration for each marker is provided in the <marker>_fold.csv file. Non-augmented training sets are available in the train_sets_no_augmentation directory. Folder structure : bouillabert_data/ ├── train_sets_no_augmentation/│ └── <marker>/│ └── fold_<n>/│ └── train_no_augment.csv│└── train_data_all_markers/ └── <marker>/ ├── <marker>_fold.csv └── fold_<n>/ ├── train.csv ├── val.csv └── test.csv Where : <marker> ∈ {teleo, mifish, berry, ac16s} <n> ∈ {1, 2, 3, 4, 5, 6}