Collect — RF3

Prepare sequences


uv run python -m data_pipelines.fasta_to_sources \
  --fasta data_pipelines/vfdb/VFDB_setA_pro.fas \
  --label 1 --name-prefix vf \
  --out data_pipelines/vfdb/sources.csv \
  --min-length 50 --max-length 300

Build inputs JSON


uv run python -m data_pipelines.build_inputs \
  --sources data_pipelines/vfdb/sources.csv \
  --out data_pipelines/vfdb/rf3_inputs.json \
  --model rf3

inputs.json format:


[
  {
    "name": "vf_VFG001",
    "components": [{ "seq": "MKVL...", "chain_id": "A" }]
  }
]

Collect


saffron collect model=rf3 \
  hooks=rf3_default \
  inputs=data_pipelines/vfdb/rf3_inputs.json \
  out_dir=outputs/collect/vfdb_rf3

Output: outputs/collect/vfdb_rf3/activations/activations.h5