In this study, we proposed a not fully connected deep learning model, DeepSignalingSynergy, for drug combination prediction. Compared with existing models that use a large number of chemical-structure and genomics features in densely connected layers, we built the model on a small set of cancer signaling pathways, which can mimic the integration of multi-omics data and drug target/mechanism in a more biological meaningful and explainable manner.

• python 3.7.3
• tensorflow 1.13.1
• pandas
• sklearn

This study intergrates following datasets

• NCI ALMANAC drug combination screening dataset
• Gene expression data of NCI-60 Cancer Cell Lines
• KEGG signaling pathways and cellular process
• Drug-Target interactions from DrugBank database

Finally, those datasets files will be parsed into numpy files to train our DeepSignalingSynergy model.

python3 parse_file.py
python3 load_data.py

Run the code

python3 main.py

Analyze the experiment results and plot figures

python3 analysis.py