Collect — RFD3

Get structures


# SafeProtein hazards (429 structures)
./data_pipelines/download_safeprotein.sh
 
# UniProt benigns with matched PDBs
./data_pipelines/download_uniprot_pdb_benigns.sh 500 100 300

Build sources + filter


# Merge CSVs, filter residues ≤300, balance per length bucket
uv run python -m data_pipelines.filter_pdbs \
  --sources tutorials/sae_data_rfd3_partial/sources.csv \
  --out tutorials/sae_data_rfd3_partial/sources_filtered.csv \
  --bin-size 50
 
uv run python -m data_pipelines.build_inputs \
  --sources tutorials/sae_data_rfd3_partial/sources_filtered.csv \
  --out tutorials/sae_data_rfd3_partial/train_inputs.json \
  --model rfd3 --partial-t 5.0

inputs.json format:


{
  "hazard_P00626": { "input": "/path/to/3g8g.pdb", "partial_t": 5.0 },
  "benign_A0PK11": { "input": "/path/to/2kw3.pdb", "partial_t": 5.0 }
}

Collect


saffron collect model=rfd3 \
  hooks=rfd3_partial \
  inputs=tutorials/sae_data_rfd3_partial/train_inputs.json \
  out_dir=outputs/collect/rfd3_train

For sequences without PDBs, first fold with RF3: rf3 fold inputs=rf3_inputs.json out_dir=pdbs/ skip_existing=True, then run attach_pdbs and optionally filter_pdbs.

Output: outputs/collect/rfd3_train/activations/activations.h5