This repository records all the models utilized in the paper, including
- Baseline model
- SAGEConv-configured GNN model
- GATv2Conv-configured GNN model
- GINConv-configured GNN model
- TransformerConv-configured GNN model
- GENConv-configured GNN model (SynerGNet)
- pytorch 1.10.0
- torch_geometric 2.0.2
- numpy 1.19.2
- sklearn 0.23.2
- pandas 1.1.3
- CUDA 11.1
- Matplotlib 3.5.1
The trained models for different configurations are recorded in ./Trained_models/
directory.
- Prepare a .csv file containing synergy instances, following the format exemplified in
./Example_data/drugcombs_synergy_data.csv
. - Transform your graph data into the .h5 format. Refer to the examples of the .h5 files located in
./Dataset/DrugCombDB/h5py_synergy_data/
for guidance.
Run python Train.py synergy_file_path h5py_dir_path model_name
synergy_file_path
represents the file path to the .csv file containing synergy instances.
h5py_dir_path
denotes the directory path where the .h5 format graphs are stored.
model_name
specifies the model that you want to run. The available options are Baseline_model
, SAGEConv
, GATv2Conv
, GINConv
, TransformerConv
, and GENConv
.
./Results_reproduction/
directory offers the reproduction of the results (figures) presented in the paper.
GENConv-configured GNN model was ultimately selected as the final model, designated as SynerGNet.
For further details on SynerGNet and instructions on its usage, please refer to https://github.com/MengLiu90/SynerGNet.
Two synergy datasets were utilized in this study:
-
Original synergy dataset
The synergy data from AZ-DREAM challenge (https://www.synapse.org/#!Synapse:syn4231880/wiki/)
-
Augmented synergy dataset
Augmented synergy data generated from the AZ-DREAM challenge synergy data. The complete augmented data can be accessed through https://osf.io/kd9e7/.