Model card

Model description

This model has been trained on varKodes. These are images encoding the genomic landscape of a species. For more information, visit the github of the varKoder project

The model architecture follows timm/vit_large_patch32_224.orig_in21k but initialized with random weights with vision_learner() function in the fastai library.

Intended uses & limitations

Since this is trained on highly specialized technical images, it is only intended for the purposes of genetic sequence identification.

Training and evaluation data

The current iteration of the model has been trained to recognize 861 eukaryotic families as described in this preprint:

de Medeiros BAS et al. 2024. A universal DNA barcode for the Tree of Life. ecoevoRxiv https://doi.org/10.32942/X24891

Outputs are multi-label predictions for each of the included families, with the following format family:[NCBI Taxonomic ID]. Check NCBI Taxonomy to translate taxonomic IDs into taxon names.