There are codes, processed data and trained models for submission paper "Predicting Protein-Ligand Binding Affinity via Joint Global-Local Interaction Modeling" in ICDM 2022.
Limited to the size of attachments, we only upload PDBbind 2016 core dataset and CSAR-HiQ dataset to test the trained models.
- PDBbind v2016 core set: ./data/2016_core_data
- CSAR-HiQ (including set1, set2): ./data/CSAR_HiQ_data
- The code for our GLI model in ./models/global_local_interaction_model.py.
- The code for related baselines models in ./models/baseline_models.py
- The code for evaluating and test in ./test_model.py
- cuda: 11.0
- GPU: V100
- Packages: The required python package are listed in requirements.txt
In our submission, we used 10 fold-cross validation to evaluate our GLI framework performance. We provide some trained models in ./trained_models which were trained in PDBbind v2016 refined dataset, and tested in PDBbind v2016 core dataset and CASR-HiQ dataset.
To indicate the GLI framework based on different models:
- GLI-0: Taking GAT+GCN as the model in chemical info embedding module.
- GLI-1: Taking GIN as the model in chemical info embedding module.
- GLI-2: Taking GCN2 as the model in chemical info embedding module.
To indicate the GLI framework with different modules:
- GLI-*-c: Including chemical info embedding module.
- GLI-*-cg: Including chemical info embedding module, global interaction module.
- GLI-*-cl: Including chemical info embedding module, local interaction module.
- GLI-*-cgl: Including chemical info embedding module, global interaction module, local interaction module.
Command:
python test_model.py GLI-0-c
python test_model.py GLI-1-c
python test_model.py GLI-2-c
Command:
python test_model.py GLI-0-cg
python test_model.py GLI-1-cg
python test_model.py GLI-2-cg
Command:
python test_model.py GLI-0-cl
python test_model.py GLI-1-cl
python test_model.py GLI-2-cl
Command:
python test_model.py GLI-0-cgl
python test_model.py GLI-1-cgl
python test_model.py GLI-2-cgl