TransCDR: a deep learning model for enhancing the generalizability of cancer drug response prediction through transfer learning and multimodal data fusion for drug representation
TransCDR is a Python implementation of a deep learning model for enhancing the generalizability of cancer drug response prediction through transfer learning and multimodal data fusion for drug representation.
TransCDR achieves state-of-the-art results of predicting CDRs in various scenarios. More importantly, TransCDR is shown to be effective in the external dataset: CCLE. In summary, TransCDR could be a powerful tool for predicting cancer drug response.
Figure 1: The overall architecture of TransCDR is located at image/
TransCDR depends on the following packages, you must have them installed before using TransCDR.
conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=10.2 -c pytorch
conda install -c dglteam/label/cu102 dgl
conda install -c rdkit rdkit
pip install dgllife
pip install matplotlib
pip install seaborn
pip install lifelines
pip install prettytable
pip install pubchempy
pip install fitlog
All datasets used in the project are located at https://zenodo.org/deposit/new. You shoud download and unzip the result.7z and put it in the current directory.
The script of data segmentation strategies for CV10 is located at folder script/
cd script
$ bash Step1_data_split.sh
train the TransCDR under various scenarios
$ bash Step2_TransCDR_CV10.sh
get the CV10 results
$ bash Step3_CV10_result.sh
$ bash Step4_Train_final_model.sh
The pre-trained TransCDR models are located at https://zenodo.org/deposit/new. You shoud download and unzip the data.7z and put it in the current directory.
$ bash Step5.1_test_on_CCLE_data.sh
$ bash Step5.2_screening_drugs_for_TCGA_patients.sh
$ python Step6_CDR_prediction.py
Xiaoqiong Xia < 19111510052@fudan.edu.cn >
Department of the Institutes of Biomedical Sciences at Fudan University.