Pancancer survival prediction using a deep learning architecture with multimodal representation and integration

Data

Please download the data from Genomic Data Commons (GDC) and University of California Santa Cruz (UCSC) Xena. Make sure the original multimodal datasets are in the 'MultimodalSurvivalPrediction/preprocess/data' directory before you start running preprocessings and experiments.

Or you can download the preprocessed data from our Google Drive and make sure the unzipped multimodal datasets are in the 'MultimodalSurvivalPrediction/preprocess/preprocessed_data' directory.

Install packages

pip install -r requirements.txt

Experiments

I. If you’d like to run experiment with specified modality combination (e.g. clinical, miRNA, and mRNA) for pancancer survival prediction, you first need to modify the 'modalities_list' variable in 'pancancer_prediction.py'.

modalities_list = [['clinical', 'mRNA', 'miRNA']]

Then, to run the experiment of your choice, simply type:

>> cd <Path_to_MultimodalSurvivalPrediction_folder>
>> python3 pancancer_prediction.py

II. If you’d like to run experiment with specified modality combination (e.g. clinical, miRNA, and mRNA) with pancancer training dataset for single cancer survival prediction, you first need to modify the 'modalities_list' variable in 'single_cancer_prediction.py'.

modalities_list = [['clinical', 'mRNA', 'miRNA']]

And then type:

>> cd <Path_to_MultimodalSurvivalPrediction_folder>
>> python3 single_cancer_prediction.py

Reference

If you find our work useful in your research, please consider citing:

@article{fan2023pancancer,
    author = {Fan, Ziling and Jiang, Zhangqi and Liang, Hengyu and Han, Chao},
    title = {Pancancer survival prediction using a deep learning architecture with multimodal representation and integration},
    journal = {Bioinformatics Advances},
    volume = {3},
    number = {1},
    pages = {1-9},
    year = {2023},
    doi = {10.1093/bioadv/vbad006}
}