xdong97/cipher-sc

CIPHER-SC

CIPHER-SC is the latest version of CIPHER. It is a complete end-to-end prediction algorithm based on a context-aware network including single-cell data.

Prerequisite

Cipher-SC is implemented mainly based on PyTorch and PyTorch Geometric (PyG). So please install these two libraries first.

• PyTorch Installation
• PyG installation

Preprocess

To train and test with CIPHER-SC, we should run preprocess first. The first step is to run generate_edgelist.py under preprocess/edgelist_result. Then follow the step 1-4 in process folder.

After that, context-aware networks and the corresponding training data will be generated under dataset folder. Dataset can be downloaded here. You should unzip and place it under the dataset folder, i.e., dataset/union_SMH_PG.

Train

For training Cipher-SC with default setting, you can directly run as follows:

python train.py 

More parameter choices can be found in train.py.

Test

We provide the checkpoint of Cipher-SC under the checkpoint folder, directly run the following code:

python test.py

Then the final result (0.9501 of AUC) in our paper is obtained.