SumGNN-DTI - Drug Target Interaction Prediction with Knowledge Graph

External Knowledge Graph Dataset - Hetionet

HETIONET

data/hetio/metaedge_encoding.json : Only the encoding value included in the original relations_2hop.txt exists

Dataset

DRUGBANK

data/drugbank/train.txt
data/drugbank/dev.txt
data/drugbank/test.txt
data/drugbank/drugbank.txt

Drug embedding - DDI

SumGNN provided

data/drugbank/DB_molecular_feats.pkl

DAVIS

Original

data/davis/Davis_train_origin.csv
data/davis/Davis_val_origin.csv
data/davis/Davis_test_origin.csv
data/davis/davis : 위의 세 파일을 단순 concat

Only data that can be used with subgraph

data/davis/Davis_train_test_val.ipynb : network filtering

data/davis/train.txt
data/davis/dev.txt
data/davis/test.txt
data/davis/davis.txt : input for hetio extractor

Drug/Target embedding

data/davis/get_dt_pkl.ipynb (코드 리팩토링 중)

data/davis/DAVIS_drug_feats.pkl
data/davis/DAVIS_target_feats.pkl

KIBA

Original

data/kiba/kiba.pkl

Only data that can be used with subgraph

data/kiba/train.txt
data/kiba/dev.txt
data/kiba/test.txt
data/kiba/kiba.txt : input for hetio extractor

Baseline models

ML Baseline - SVM, XGBoost, Randomforest

BaselineModel/train_baseline_model.ipynb

python baselinemodel.py 
    -td ./data/davis/Davis_test_origin.csv    # Dataset for Training  
    -vd ./data/davis/Davis_val_origin.csv     # Dataset for Validation  
    -m SVM     #(SVM, XGBoost, RandomForest)  # Choose model to train and see the result. 

DL Baseline

HyperAttentionDTI

Train

bash run.sh ->

python train.py -d davis -e ddi_hop2 --gpu=2 --hop=2 --batch=128 -b=4 --num_epochs 500 --lr 0.00005