Skip to content

Latest commit

 

History

History
56 lines (41 loc) · 4.81 KB

GEMS_pdbbind.md

File metadata and controls

56 lines (41 loc) · 4.81 KB

Preprocessing and Graph Dataset Construction

Download the PDBbind database from http://www.pdbbind.org.cn/. Then follow the steps below to construct a dataset of affinity-labelled interactions graphs and run trainining/inference. At least 100GB of storage are needed for preprocessing 20'000 protein-ligand complexes.

  • Ligand Preprocessing: As a first step in preprocessing the ligand files, we have used the mol2 files in PDBbind and converted them into SDF files with explicit hydrogens. The resulting SDF files were then used for the graph construction below.

    obabel -imol2 <input/mol2> -osdf -O <output/SDF> -h

  • Prepare your data: Save all PDB and the SDF files (including CASF complexes) in the same directory. Each protein-ligand pair should share the same unique identifier (ID) as filenames to indicate they form a complex. For example, use filenames like ID.pdb and ID.sdf to represent the same complex. As a first step in the preprocessing, we have added explicit hydrogens to all ligand SDF files using obabel:

  • Prepare the labels: Use the provided PDBbind data dictionary (PDBbind_data/PDBbind_data_dict.json) or parse the index file of PDBbind into a json dictionary (you can use PDBbind_data/read_index_into_dict.py, but you might have to adjust some paths)

  • Compute Language Model Embeddings: To compute ChemBERTa-77M, ANKH-Base and ESM2-T6 embeddings and save them in your data directory, execute the following commands:

    ChemBERTa: python -m dataprep.chemberta_features --data_dir <path/to/data/dir> --model ChemBERTa-77M-MLM
    ANKH: python -m dataprep.ankh_features --data_dir <path/to/data/dir> --ankh_base True
    ESM2: python -m dataprep.esm_features --data_dir <path/to/data/dir> --esm_checkpoint t6

    You can also include more or only a subset of these embeddings or change to ChemBERTa-10M (--model ChemBERTa-10M-MLM), to ANKH-Large (--ankh_base False) or to ESM2-T33 (--esm_checkpoint t33). We recommend running these scripts on a GPU.

  • Run the graph construction: To construct interaction graphs for all protein-ligand complexes in your data directory (incorporating language model embeddings), run the following command with the desired combination of protein and ligand embeddings:

    python -m dataprep.graph_construction --data_dir <data/dir> --protein_embeddings ankh_base esm2_t6 --ligand_embeddings ChemBERTa_77M

  • Run the dataset construction: You need to provide the path to the directory containing your data (--data_dir) and the path to save the dataset (--save_path). To include the labels, provide also the path to the JSON file containing the log_kd_ki values. In addition, provide the split dictionary and the dataset (corresponding to a key in the dictionary) for which the PyTorch dataset should be contstructed. Finally, add the protein embeddings and the ligand embeddings that should be used to featurize the graphs (any combination of the embeddings included in the graph construction process is possible). This will generate a pytorch dataset of affinity-labelled interactions graphs featurized with the desired language model embeddings.

    PDBbind training dataset: Replace "train" with "casf2013" or "casf2016" to build datasets for CASF

    python -m dataprep.construct_dataset --data_dir <data/dir> --save_path <output/path.pt> --data_split PDBbind_data/PDBbind_data_split_pdbbind.json --dataset train --data_dict PDBbind_data/PDBbind_data_dict.json --protein_embeddings ankh_base esm2_t6 --ligand_embeddings ChemBERTa_77M

    PDBbind CleanSplit training dataset: Replace "train" with "casf2013" or "casf2016" to build datasets for CASF

    python -m dataprep.construct_dataset --data_dir <data/dir> --save_path <output/path.pt> --data_split PDBbind_data/PDBbind_data_split_cleansplit.json --dataset train --data_dict PDBbind_data/PDBbind_data_dict.json --protein_embeddings ankh_base esm2_t6 --ligand_embeddings ChemBERTa_77M

Run GEMS on the generated datasets

  • Inference: You can now run inference or training on the generated PyTorch datasets. This command will automatically run inference using the model appropriate for the dataset type (if there is one for the chosen combination of embeddings): 
    python inference.py --dataset_path <path/to/dataset>
  • Training: 
    python train.py --dataset_path <path/to/dataset_file> --run_name <select a run name>
  • Test:
    Test the newly trained model with test.py, using the saved stdict and the path to a test dataset as input. If you want to test an ensemble of several models, provide all stdicts in a comma-separated string. 
    python test.py --dataset_path <path/to/downloaded/test/set> --stdicts <path/to/saved/stdict>