Transductive learning from a transformer foundation model improves the taxonomic classification of eDNA metabarcodes: Code and data

Fish Marker Family Assignment Benchmark This archive contains the code and data used for the benchmark study on fish marker family assignment, as described in the paper "Transductive learning from a transformer foundation model improves the taxonomic classification of eDNA metabarcodes". Data For this study, we utilized the MIDORI2 Reference and MitoFish databases. Teleo, Mifish, Berry and Ac16 markers extraction was performed using the CRABS tool via Docker execution (details provided in data/README.md) Methods Detailed information about the requirements and execution of each script is available in their respective directories under `scripts/`. All methods were tested on a Linux operating system. For use on other platforms, please refer to the original documentation provided by the authors (links below). - OBITOOLS 4: code - MMSEQS2: code/paper - BERTAX: code/paper - BouillaBert: code/paper - DNAGPT: code/paper Results To enable figure reproduction, the prediction csv file (results/all_predictions_pre_compute.csv) is provided along with the corresponding scripts (results/figure_scripts.py).