conda create --name antifold python=3.10 -y && conda activate antifold
conda install -c conda-forge pytorch
git clone https://github.com/oxpig/AntiFold && cd AntiFold
pip install .

conda create env -f environment.yml
python -m pip install .

# Run AntiFold on single PDB/CIF file
# Nb: Assumes first chain heavy, second chain light
python antifold/main.py \
    --pdb_file data/pdbs/6y1l_imgt.pdb

# Antibody-antigen complex
python antifold/main.py \
    --pdb_file data/antibody_antigen/3hfm.pdb \
    --heavy_chain H \
    --light_chain L \
    --antigen_chain Y

# Nanobody or single-chain
python antifold/main.py \
    --pdb_file data/nanobody/8oi2_imgt.pdb \
    --nanobody_chain B

# Folder of PDB/CIFs
# Nb: Assumes first chain heavy, second light
python antifold/main.py \
    --pdb_dir data/pdbs

# Specify chains to run in a CSV file (e.g. antibody-antigen complex)
python antifold/main.py \
    --pdb_dir data/antibody_antigen \
    --pdbs_csv data/antibody_antigen.csv

# Sample sequences 10x (paired VH/VL only)
python antifold/main.py \
    --pdb_file data/pdbs/6y1l_imgt.pdb \
    --heavy_chain H \
    --light_chain L \
    --num_seq_per_target 10 \
    --sampling_temp "0.2" \
    --regions "CDR1 CDR2 CDR3"

# Run all chains with ESM-IF1 model weights
python antifold/main.py \
    --pdb_dir data/pdbs \
    --esm_if1_mode

usage:
    # Predict on example PDBs in folder
python antifold/main.py \
    --pdb_file data/antibody_antigen/3hfm.pdb \
    --heavy_chain H \
    --light_chain L \
    --antigen_chain Y # Optional

Predict inverse folding probabilities for antibody variable domain, and sample sequences with maintained fold.
PDB structures should be IMGT-numbered, paired heavy and light chain variable domains (positions 1-128).

For IMGT numbering PDBs use SAbDab or https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/anarci/

options:
  -h, --help            show this help message and exit
  --pdb_file PDB_FILE   Input PDB file (for single PDB predictions)
  --heavy_chain HEAVY_CHAIN
                        Ab heavy chain (for single PDB predictions)
  --light_chain LIGHT_CHAIN
                        Ab light chain (for single PDB predictions)
  --antigen_chain ANTIGEN_CHAIN
                        Antigen chain (optional)
  --pdbs_csv PDBS_CSV   Input CSV file with PDB names and H/L chains (multi-PDB predictions)
  --pdb_dir PDB_DIR     Directory with input PDB files (multi-PDB predictions)
  --out_dir OUT_DIR     Output directory
  --regions REGIONS     Space-separated regions to mutate. Default 'CDR1 CDR2 CDR3H'
  --num_seq_per_target NUM_SEQ_PER_TARGET
                        Number of sequences to sample from each antibody PDB (default 0)
  --sampling_temp SAMPLING_TEMP
                        A string of temperatures e.g. '0.20 0.25 0.50' (default 0.20). Sampling temperature for amino acids. Suggested values 0.10, 0.15, 0.20, 0.25, 0.30. Higher values will lead to more diversity.
  --limit_variation     Limit variation to as many mutations as expected from temperature sampling
  --extract_embeddings  Extract per-residue embeddings from AntiFold / ESM-IF1
  --custom_chain_mode   Run all specified chains (for antibody-antigen complexes or any combination of chains)
  --exclude_heavy       Exclude heavy chain from sampling
  --exclude_light       Exclude light chain from sampling
  --batch_size BATCH_SIZE
                        Batch-size to use
  --num_threads NUM_THREADS
                        Number of CPU threads to use for parallel processing (0 = all available)
  --seed SEED           Seed for reproducibility
  --model_path MODEL_PATH
                        Alternative model weights (default models/model.pt). See --use_esm_if1_weights flag to use ESM-IF1 weights instead of AntiFold
  --esm_if1_mode        Use ESM-IF1 weights instead of AntiFold
  --verbose VERBOSE     Verbose printing