Skip to content

Latest commit

 

History

History
91 lines (69 loc) · 4.4 KB

dti.md

File metadata and controls

91 lines (69 loc) · 4.4 KB
Raw

Drug-Target Interaction Prediction

Drug-target interaction prediction aims to predict the binding relationships between drugs and target proteins, which is a significant task in AI drug discovery.

Features

  • Supported models: DeepDTA, MGraphDTA and DeepEIK. This is a continuing effort and we are working on further growing the list.
  • Supported datasets: 2 classification datasets i.e. Yamanishi08's and BMKG-DTI, 2 regression datasets i.e. Davis and KIBA.
  • Supported split: warm, cold-drug, cold-protein, cold-cluster;
  • Supported evaluation: ROC_AUC, PR_AUC, F1, Precision and Recall for classification; MSE, Pearson coefficient, Spearman coefficient, CI index and $r_m^2$ index for regression.

Additional Packages

To get full access to DTI functions (all splitting strategies and evaluation metrics), run the following:

# NOTE: Make sure you are at OpenBioMed/ directory
pip install lifelines

# NOTE: ONLY RUN THE FOLLOWING SECTION IF YOU HAVE NOT YET INSTALLED
# PyBioMed
#-------------------------------------------------------------------
cd assets # create this directory if not yet existent
# PyBioMed for calculating protein descriptors
git clone https://github.com/gadsbyfly/PyBioMed
cd PyBioMed
python setup.py install
cd ..
#-------------------------------------------------------------------

To reproduce DeepEIK, install cogdl for calculating network embeddings:

git clone https://github.com/THUDM/cogdl
cd cogdl
pip install -e .

Data Preparation

Davis and KIBA can download here, Yamanishi08 and BMKG_DTI can download from here. The 4 datasets should put under datasets/dti/. To reproduce DeepEIK, install BMKGv1 and put it under assets/kg/

Model Preparation

To reproduce DeepEIK, you should install PubMedBERT from huggingface and put the checkpoint under ckpts/text_ckpts/.

Training and Evaluation

You can run scripts using bash under scripts/aidd/dti/:

scripts/aidd/dti
├── train_baseline.sh             # running DeepDTA, MGraphDTA or MolFM on Yamanishi08's and BMKG-DTI under 4 settings
├── train_baseline_regression.sh  # running DeepDTA, MGraphDTA or MolFM on Davis or KIBA
├── train_deepeik.sh              # running DeepEIK on Yamanishi08's and BMKG-DTI under 4 settings
└── train_deepeik_regression.sh   # running DeepEIK on Davis or KIBA

Example:

bash scripts/aidd/dti/train_baseline.sh cuda:0 # switch to your own cuda device or cpu

You can also modify the scripts or directly use the following command:

python open_biomed/tasks/mol_task/dti.py \
--device DEVICE \                   # gpu device id
--mode MODE \                       # training mode, kfold: 5-fold validation, train: train-test
--config_path CONFIG_PATH \         # configuration file, see configs/dti/ for more details
--dataset DATASET \                 # dataset name, select from [yamanishi08, bmkg-dti, davis, kiba]
--dataset_path DATASET_PATH \       # path to the dataset
--split_strategy SPLIT_STRATEGY \   # split strategy, select from [warm, cold-drug, cold-protein, cold-cluster]
--init_checkpoint INIT_CHECKPOINT \ # checkpoint path used for efficient validation
--param_key PARAM_KEY \             # key of the checkpoint dict that contains model parameters
--output_path OUTPUT_PATH \         # save checkpoint path used for training
--num_workers NUM_WORKERS \         # number of workers when loading data
--epochs EPOCHS \                   # number of training epochs
--patience PATIENCE \               # number of tolerant epochs for early-stopping
--weight_decay WEIGHT_DECAY \       # weight decay, default is 1e-5
--lr LR \                           # learning rate, default is 1e-3
--batch_size BATCH_SIZE \           # batch size, default is 128
--gradient_accumulation_steps GRADIENT_ACCUMULATION_STEPS \ # steps for gradient accumulation, default is 1
--logging_steps LOGGING_STEPS \     # steps for printing training information
--seed SEED                         # random seed