Each dataset is serialized as python pkl dictionary where keys are sequences and values are embeddings. Embeddings were produced using the protbert_bfd model from ProtTrans (DOI: 10.1109/TPAMI.2021.3095381). Example code for loading the embedding dictionaries can be found in the project github repository- https://github.com/kellylab/viral-protein-function-annotation-with-protein-language-model PHROGs sequences come from the Prokaryotic virus Remote Homologous Groups database v3 (DOI: 10.1093/nargab/lqab067) EFAM sequences come from the efam database (DOI: 10.25739/9vze-4143)