Pisces: A cross-modal contrastive learning approach to synergistic drug combination prediction

This repository is the official implementation of Pisces: A cross-modal contrastive learning approach to synergistic drug combination prediction. Our work has been accepted by RECOMB 2023. The code is originally forked from Fairseq and DVMP.

Requirements and Installation

• PyTorch version == 1.8.0
• PyTorch Geometric version == 1.6.3
• RDKit version == 2020.09.5

You can build the Dockerfile or use the docker image teslazhu/pretrainmol36:latest.

To install the code from source

git clone https://github.com/linjc16/Pisces.git

pip install fairseq
pip uninstall -y fairseq 

pip install ninja
python setup.py build_ext --inplace

Getting Started

Dataset

Refer to this file.

Data Preprocessing

We evaluate our models on the dataset above. dds/scripts/train_trans/data_process, dds/scripts/train_leave_comb/data_process and dds/scripts/train_leave_cell/data_process are folders for preprocessing of 5-fold CV, Stratified CV for drug combinations, and Stratified CV for cell lines settings respectively. To generate the binary data for fairseq, take the 5-fold CV setting (fold 0) as an example, run

python dds/scripts/train_trans/data_process/split_trans.py

bash dds/scripts/train_trans/data_process/run_process_trans.sh fold0

bash dds/scripts/train_trans/data_process/run_binarize_trans.sh

Note that you need to change the file paths accordingly. More original data can be found here.

Training and Test

All training and test scripts can be seen in dds/scripts. For instance,

bash dds/scripts/train_trans/run_dv_ppiv2_cons_tri.sh fold0 5e-5 0.01

bash dds/scripts/train_trans/inference/inf_dv_ppiv2_cons_tri.sh fold0 5e-5 0.01

Contact

Please feel free to submit a Github issue if you have any questions or find any bugs. We do not guarantee any support, but will do our best if we can help.