# Setup environment and install
conda create --name inverse python=3.9 -y
conda activate inverse
conda install -c pyg pyg -y
conda install -c conda-forge pip -y

git clone https://github.com/Magnushhoie/discotope3_web/
cd discotope3_web/
pip install .

# Unzip models to use
unzip models.zip

# 1. Predict single PDB (solved structure)
python discotope3/main.py --pdb_or_zip_file data/example_pdbs_solved/7c4s.pdb
# CPU only:
python discotope3/main.py --cpu_only --pdb_or_zip_file data/example_pdbs_solved/7c4s.pdb

# Setup environment with conda
conda create -n inverse python=3.9
conda activate inverse
conda install pytorch=1.11 cudatoolkit=11.3 -c pytorch
conda install pyg -c pyg -c conda-forge
conda install pip

# install pip dependencies
pip install .

# install pip dependencies
pip install -r requirements_recommended.txt
pip install .

# Unzip models
unzip models.zip

# Now select one of multiple options:

# 1. Predict single PDB (solved)
python discotope3/main.py --pdb_or_zip_file data/example_pdbs_solved/7c4s.pdb

# 2. Predict AlphaFold structure
python discotope3/main.py --pdb_or_zip_file data/example_pdbs_alphafold/7tdm_B.pdb --struc_type alphafold

# 3. Predict a folder of PDBs
python discotope3/main.py --pdb_dir data/example_pdbs_solved --out_dir output/example_pdbs_solved

# 4. Predict a ZIP file of PDBs
python discotope3/main.py --pdb_or_zip_file pdbs_in_zipfile.zip --out_dir output/pdbs_in_zipfile

# 5. Fetch PDBs from RCSB
python discotope3/main.py --list_file pdb_list_solved.txt --struc_type solved --out_dir output/pdb_list_solved

# 6. Fetch PDBs from Alphafolddb
python discotope3/main.py --list_file pdb_list_af2.txt --struc_type alphafold --out_dir output/pdb_list_af2

Predict B-cell epitope propensity on input protein PDB structures

optional arguments:
  -h, --help            show this help message and exit
  -f PDB_OR_ZIP_FILE, --pdb_or_zip_file PDB_OR_ZIP_FILE
                        Input file, either single PDB or compressed zip file with multiple PDBs
  --list_file LIST_FILE
                        File with PDB or Uniprot IDs, fetched from RCSB/AlphaFolddb
  --struc_type STRUC_TYPE
                        Structure type from file (solved | alphafold)
  --pdb_dir PDB_DIR     Directory with AF2 PDBs
  --out_dir OUT_DIR     Job output directory
  --models_dir MODELS_DIR
                        Path for .json files containing trained XGBoost ensemble
  --calibrated_score_epi_threshold CALIBRATED_SCORE_EPI_THRESHOLD
                        Calibrated-score threshold for epitopes [low 0.40, moderate (0.90), higher 1.50]
  --no_calibrated_normalization
                        Skip Calibrated-normalization of PDBs
  --check_existing_embeddings CHECK_EXISTING_EMBEDDINGS
                        Check for existing embeddings to load in pdb_dir
  --cpu_only            Use CPU even if GPU is available (default uses GPU if available)
  --max_gpu_pdb_length MAX_GPU_PDB_LENGTH
                        Maximum PDB length to embed on GPU (1000), otherwise CPU
  --multichain_mode     Predicts entire complexes, unsupported and untested
  --save_embeddings SAVE_EMBEDDINGS
                        Save embeddings to pdb_dir
  --web_server_mode     Flag for printing HTML output
  -v VERBOSE, --verbose VERBOSE
                        Verbose logging

python discotope3/main.py --pdb_or_zip_file data/example_pdbs_solved/7c4s.pdb

# Unzip AlphaFold2 test set
unzip data/test_set_af2.zip -d data/

# Run predictions on PDB folder
python discotope3/main.py \
--pdb_dir data/test_set_af2 \
--struc_type alphafold \
--out_dir output/test_set_af2

# Make sure gcc and g++ versions are updated, pybind11 is available
# torch-scatter should be listed with 'conda list' or 'pip list'

# With conda:
conda install -c conda-forge pybind11 gcc cxx-compiler

# With apt-get
sudo apt-get install gcc g++
pip install pybind11