# Download code
mkdir -p antifold_code && cd antifold_code
wget https://opig.stats.ox.ac.uk/data/downloads/AntiFold/antifold_code.zip
unzip antifold_code.zip

# Download model
mkdir -p models
wget -P models/ https://opig.stats.ox.ac.uk/data/downloads/AntiFold/models/model.pt

conda create --name antifold python=3.10 -y
conda activate antifold
conda install -c pytorch pytorch
conda install -c pyg pyg -y
conda install -c conda-forge pip -y
# Install AntiFold from antifold_code directory
pip install .
pip install torch-geometric==2.4.0

conda create --name antifold python=3.10 -y
conda activate antifold
conda install -c conda-forge pytorch-gpu
conda install -c pyg pyg -y
conda install -c conda-forge pip -y
# Install AntiFold from antifold_code directory
pip install .
pip install torch-geometric==2.4.0

# Residue probabilities, sample 10 sequences in CDRH3 from single PDB
python antifold/main.py \
    --out_dir output/single_pdb \
    --pdb_file data/pdbs/6y1l_imgt.pdb \
    --heavy_chain H \
    --light_chain L \
    --num_seq_per_target 10 \
    --sampling_temp "0.2" \
    --regions "CDRH3"

# Residue probabilities, sample 10 sequences at temperatures 0.20 & 0.30 in CDRs 1-3 from folder of PDBs
python antifold/main.py \
    --out_dir output/example_pdbs \
    --pdbs_csv data/example_pdbs.csv \
    --pdb_dir data/pdbs \
    --num_seq_per_target 10 \
    --sampling_temp "0.20 0.30" \
    --regions "CDR1 CDR2 CDR3"

# Extract ESM-IF1 embeddings from PDBs in folder, from all chains listed in untested.csv (not limited to VH/VL)
python antifold/main.py \
    --out_dir output/untested/ \
    --pdbs_csv data/untested.csv \
    --pdb_dir data/untested/ \
    --use_esm_if1_weights \
    --custom_chain_mode \
    --extract_embeddings

usage: 
    # Predict on example PDBs in folder
    python antifold/main.py     --pdbs_csv data/example_pdbs.csv     --pdb_dir data/pdbs     --out_dir output/

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 (experimental)
  --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   Custom chain input (experimental, e.g. single chain, inclusion of antigen chain or any chains with ESM-IF1)
  --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
                        AntiFold model weights. See --use_esm_if1_weights flag to use ESM-IF1 weights instead of AntiFold
  --use_esm_if1_weights
                        Use ESM-IF1 weights instead of AntiFold
  --verbose VERBOSE     Verbose printing