DrugRepBench: a multi-scenario benchmark for knowledge graph-based drug repurposing

This repository is the official implementation of "DrugRepBench: a multi-scenario benchmark for knowledge graph-based drug repurposing".

Overview of DrugRepBench

Requirements

To install requirements:

pip install -r requirements.txt

Data Processing

We use two KG examples to demonstrate the process of creating our benchmark. First, navigate to the directory data_process and unzip the Big_data.zip file to obtain the required data files.

cd data_process
unzip Big_data.zip

Data files obtained from Big_data.zip need to be moved to the corresponding directory according to the following path record.

benchmark/graph.txt
data_process/data/example_kg1/kg.csv
data_process/data/embeddings.json
data_process/data/kg1_processed.json
data_process/data/kg2_processed.json
data_process/kg_merged/kg-data.gpickle

Then run the 1merge.py file to merge the two example KG files.

python 1merge.py

Then, run file 2kg_build.py to build a new KG based on the merged information.

python 2kg_build.py

The test samples for Fine-grained Cold-Start Scenarios can be obtained by running file 3kg_divide.py.

python 3kg_divide.py

Running 4hard_negative.py can obtain data information for the Hard Negative Sampling Scenario, but you need to first construct the embedding of the KG node and the mechanistic difference set of the candidate drugs. Here we provide files embeddings.json and drug_mechset.json for reference.

DrugRepBench

The content of benchmark DrugRepBench is located under the directory benchmark and includes graph.txt, train.txt, valid.txt, and three test Scenarios.