Uniprot datasets for training taxonomic classification

The data is based on the UniProt-Swiss-Prot release 2020-04 dataset and contains data derived from amino acid sequences of human, bacterial and viral origin. From each original sequence we created multiple patches of length 100 using a sliding window. The data is stored in the FASTA format according to

>{ID}_{patch index}|{class marker}
sequence

with

ID - denotes the ID of the original sequence in the UniProt-Swiss-Prot dataset
sequence - patch of length 100 of an amino acid sequence
patch_index - denotes the starting index of the given patch within the original sequence
class marker - indicates the taxonomic class
    0 - virus
    1 - bacteria
    2 - human / mammal

The data is split into test, training and validation set which contain the following number of patches per class:

- train: 4.891.278
- test: 611.602
- val: 611.602