GraphormerDTI

A graph transformer-based approach for drug-target interaction prediction.

Requirements and Installation

pip install -r requirements.txt

Preprocess

Under transductive setting, please run the code

python preprocess_transductive_setting.py

Under drug inductive setting, please run the code

python preprocess_drug_inductive_setting.py

Under drug-protein inductive setting, please run the code

python preprocess_drug_protein_inductive_setting.py

Using the Davis dataset under transductive setting as an example, the preprocessing might output the following log

Train in Davis
proteins load finished
data shuffle
10157 2540 12696
10157 finished
2540 finished
12696 finished
train dataset finished
valid dataset finished
test dataset finished
10157 2540 12696

The logs show that under the current random seed, the training set, validation set and test set include 10157, 2540 and 12696 drug-protein pairs, respectively. The pre-processing result is stored in ./molecule_data/Davis_transductive_setting_SEED.pkl.

Run

Under transductive setting, please run the code

python main_transductive_setting.py

Under drug inductive setting, please run the code

python main_drug_inductive_setting.py

Under drug-protein inductive setting, please run the code

python main_drug_protein_inductive_setting.py

We will continue to use the Davis dataset under transductive setting as an example, using ./protein_data/Davis.txt and the preprocessed output pkl file as inputs for model training and prediction. During the process, logs may be output as follows:

Train in Davis
proteins load finished
data shuffle
10157 2540 12696
train, val, test sizes : 10157 2540 12696
molecules load finished
Training...
The stable model's results:...

The log output includes the results of the validation set after each epoch and the final test set results, which are stored in ./Davis_transductive_setting folder.