UniProt search (6/17/2025): 
(cc_scl_term_exp:SL-0086) OR (cc_scl_term_exp:SL-0191) OR (cc_scl_term_exp:SL-0243) OR (cc_scl_term_exp:SL-0162) OR (cc_scl_term_exp:SL-0173) OR (cc_scl_term_exp:SL-0209) OR (cc_scl_term_exp:SL-0095) OR (cc_scl_term_exp:SL-0272) OR (cc_scl_term_exp:SL-0158) OR (cc_scl_term_exp:SL-0132) OR (cc_scl_term_exp:SL-0204) OR (cc_scl_term_exp:SL-0049) OR (cc_scl_term_exp:SL-0041)

Filtered for specific subcellular location terms: 
(1) Cytoplasm
(2) Nucleus
(3) Secreted
(4) Membrane
(5) Mitochondria
(6) Plastid
(7) Endoplasmic Reticulum
(8) Vacuole
(9) Lysosome
(10) Golgi Apparatus
(11) Peroxisome
(12) Chloroplast
(13) Cell Wall

Curation steps:
- Filtered for sequence length >=20 and <=2048
- CD-HIT using Docker at 80% similarity threshold and n=5
- Keep representative sequences
- Parse subcellular locations into a clean list of labels (remove notes & ECO numbers)
- Keep only examples with label in the location map (above)
- Remove duplicates
- Unique label list (assign 0 or 1 to a bitstring), convert to vector
- Shuffle and split randomly into train, valid (1k), and test (1k)

---
dataset_info:
  features:
  - name: labels
    dtype: string
  - name: seqs
    dtype: string
  splits:
  - name: train
    num_bytes: 15913904
    num_examples: 32638
  - name: valid
    num_bytes: 472742
    num_examples: 1000
  - name: test
    num_bytes: 484576
    num_examples: 1000
  download_size: 15433449
  dataset_size: 16871222
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: valid
    path: data/valid-*
  - split: test
    path: data/test-*
---