Hi-GeoMVP is a deep-leaning model using graph neural networks to represent the drug and cell line features for drug response prediction.

Hi-GeoMVP takes drug geometry, chemical graph, and fingerprints, along with multi-omics data as input. Hi-GeoMVP employs a geometry-enhanced graph neural network (GeoGNN) and an multi-layer perceptron (MLP) for drug representation learning from drug features, and graph attention networks (GATa) and a variational autoencoder (VAE) for cell representation learning from multi-omics data. These representations are then integrated via a multi-view fusion module for drug response prediction (ln(IC50)). The model also utilizes latent representations alongside distinct prediction sub-networks for joint learning of drug and cell tasks, ultimately improving the prediction of drug response. The final output of the Hi-GeoMVP model is the predicted value of ln(IC50).

Run create_cell_feat.py to generate the used cell_line feature.
Run create_drug_feat.py to generate the used drug feature. 
Run create_drp_dict.py to generate the drug_cell_ic50 feature and train_test split for mix,cell_blind and drug_blind sets.
All the used data are saved in .npy format and under the ./Data/DRP_dataset
An example to create the dataset and dataloader in Pytorch is provided in data_preprocess.py

We provide the script to train the model with different settings and hyperparameters:
train_type: 'cb', 'db', or 'mix' for cell blind, drug blind, and mix setting.
use_norm_ic50: 'True' or 'False', for usage of normalized IC50 value.
use_regulizer: 'True' or 'False', usage of MTL cancer type.
use_regulizer_drug: 'True' or 'False', usage of MTL drug threshold.
use_drug_path_way: 'True' or 'False', usage of MTL drug targeting pathway.
use_raw_gene: 'True' or 'False', usage of whole gene expression data. regular_weight: float number, the weight of MTL cancer type in the loss function.
regular_weight_drug: float number, the weight of MTL drug threshold in the loss function.
regular_weight_drug_path_way: float number, the weight of drug targeting pathway in the loss function.

You can run the training in the command line with:

python train_GeoMVP_5fold.py --train_type {train_type} 
    --device {device}--use_norm_ic50' {use_norm_ic50}  
    --use_drug_path_way {use_drug_path_way} 
    --use_regulizer {use_regulizer}
    --use_regulizer_drug {use_regulizer_drug} 
    --use_raw_gen {use_raw_gen}