metadata
dataset_info:
features:
- name: seqs
dtype: string
- name: cluster_ids
dtype: string
- name: labels
dtype: int64
splits:
- name: train
num_bytes: 130010008
num_examples: 248656
- name: valid
num_bytes: 7068366
num_examples: 13766
- name: test
num_bytes: 7219205
num_examples: 13807
download_size: 120481123
dataset_size: 144297579
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: valid
path: data/valid-*
- split: test
path: data/test-*
- All Uniprot entries 7/1/2025 with manual GO annotation
- Keep entries with only one GO CC
- Only keep labels with at least 10000 examples
- cluster at 50% with cd-hit
- split valid and test randomly by assigning 5% of the clusters each
- sample a maximum of 10 samples from each cluster randomy to form the training, valid, and test split
GO terms mapping: 0: extracellular region [GO:0005576] 1: membrane [GO:0016020] 2: plasma membrane [GO:0005886] 3: cytoplasm [GO:0005737] 4: endoplasmic reticulum membrane [GO:0005789] 5: cytosol [GO:0005829] 6: mitochondrion [GO:0005739] 7: extracellular space [GO:0005615] 8: nucleus [GO:0005634]