SWnet: a deep learning model for drug response prediction from cancer genomic signatures and compound chemical structures

Core code for the paper "SWnet: a deep learning model for drug response prediction from cancer genomic signatures and compound chemical structures" by Zhaorui Zuo, Penglei Wang, Xiaowei Chen, Li Tian, Hui Ge & Dahong Qian.

Resources

---

Data

The data in the folder is prepared for training and evaluating the SWnet.

• data/GDSC/drug_similarity/GDSC_drug_similarity.csv: This csv file record the similarity of drugs.
• data/GDSC/GDSC_data: The GDSC data which include 1478 genes across 1018 cell lines.
• data/GDSC/graph_data: The molecular graph information is saved in this data file.
• data/CCLE/drug_similarity/CCLE_drug_similarity.csv: This csv file record the similarity of drugs.
• data/CCLE/CCLE_data: The CCLE data which include 1478 genes across 469 cell lines.
• data/CCLE/graph_data: The molecular graph information is saved in this data file.

Installation

---

Install the requirements (listed in environment.yaml). We're using Anaconda to install the environment:

conda create -f environment.yaml
conda activate swnet
pip install numpy==1.16.2

Running the Code

---

Model Code

As shown below, SWnet adopts a dual converge architercture.Genomic signature and chemical fingerprints are porcessed in parallel through GNN and CNN layers to extract independent features, which are then concatenated. And SWnet also integrate multi-task learning and self-attentation mechanism to further improve the performance. The code for the SWnet can be found in multi-task, self-attention, single-layer.

Evaluation on pretrained model

• cd self-attention
• python SWnet_GDSC_self-attention_evaluate.py
• python SWnet_CCLE_self-attention_evaluate.py

or

Train a prediction model on GDSC data

Prepare graph data, we can set the radius parameter to 1, 2, 3 or 4

• cd data/GDSC
• python preprocess_drug_graph.py --radius 1

Prepare drug similarity data

• cd data/GDSC
• python preprocess_drug_similarity.py

Train self-attention SWnet

• cd self-attention
• python SWnet_GDSC_self-attention_train.py

you can set hyper-parameter like this:

• python SWnet_GDSC_self-attention_train.py --radius 3 --split_case 0 --layer_gnn 3

Evaluate self-attention SWnet

• cd self-attention
• python SWnet_GDSC_self-attention_evaluate.py

or

Train a prediction model on CCLE data

Prepare graph data, we can set the radius parameter to 1, 2 ,3 or 4

• cd data/CCLE
• python preprocess_drug_graph.py --radius 1

Prepare drug similarity data

• cd data/CCLE
• python preprocess_drug_similarity.py

Train self-attention SWnet

• cd self-attention
• python SWnet_CCLE_self-attention_train.py

you can set hyper-parameter like this:

• python SWnet_CCLE_self-attention_train.py --radius 3 --split_case 0 --layer_gnn 3

Evaluate self-attention SWnet

• cd self-attention
• python SWnet_CCLE_self-attention_evaluate.py

Run Other scripts

The following scripts training the muti-task SWnet.

• cd multi-task
• python SWnet_multi-task.py

The following scripts training the single-layer SWnet.

• cd single-layer
• python SWnet_single_no_weight.py
• python SWnet_single_yes_weight.py

The following scripts training the GDSC gene weight Layer.

• cd self-attention
• python SWnet_GDSC_self-attention_train.py --radius 3 --split_case 0
• python SWnet_CCLE_self-attention_train.py --radius 3 --split_case 0

Citation

---

If you find this code useful for your research, please use the following citation.

Zuo, Z., Wang, P., Chen, X. et al. SWnet: a deep learning model for drug response prediction from cancer genomic signatures and compound chemical structures. BMC Bioinformatics 22, 434 (2021). https://doi.org/10.1186/s12859-021-04352-9