Repository for Multimodal Network Diffusion Predicts Future Disease-Gene-Chemical Associations.
If you have any questions or comments, feel free to contact Chih-Hsu Lin (chihhsul[at]bcm[dot]edu).

• Download code
• Installation and download network data
• Run tutorial
• Run experiments
• Project organization
• Links to download data

git clone https://github.com/LichtargeLab/multimodal-network-diffusion

• Anaconda or MiniConda

• For Mac: Julia 0.5.2

• For Linux: Julia 0.6.3

• Add Julia path in system variable PATH Mac commands are below. You might need to change the Julia path if you install in a different location.

echo "export PATH=\"/Applications/Julia-0.5.app/Contents/Resources/julia/bin:\$PATH\"" >> "$HOME/.bash_profile" 
source ~/.bash_profile

cd multimodal-network-diffusion
./install.sh

Mapped network data and the mapping file will be automatically downloaded and extracted to ./data/mapped_networks/.

cd notebooks/
jupyter notebook

The kernel name could be Python [Diffusion2018] or Python [conda env:Diffusion2018] in our tests.

source activate Diffusion2018

python -m ipykernel install --user --name Diffusion2018 --display-name "Python [Diffusion2018]"

./download_precomp.sh

Enter y or n for downloading experiments of interests:

Please choose experiments of interests:
10-fold cross-validation (275 GB after compression) [y]/n ? 
Leave-one-mode-out (9.4 GB after compression) [y]/n ? 
Time-stamped (22 GB after compression) [y]/n ? 
Prospective (21.6 GB after compression) [y]/n ? 

source activate Diffusion2018

If it's activated, you will see (Diffusion2018) at the beginning of your command prompt

Experiment parameter: 1mode:

cd src
python runKfoldCV.py --e 1mode # Could be one of {'1mode','3mode','6mode'}

Experiment parameter options: 1mode: 1-mode networks; 3mode: 3-mode networks; 6mode: 6-mode networks

cd src
python runLeaveAModeOut.py --e DG # Could be one of {'DG','DC','GC'}

Experiment parameter options: DG: disease-gene; DC: disease-chemical; GC: gene-chemical

cd src
python runTimeStamped.py --e DC # Could be one of {'DG','DC','GC'}

Experiment parameter options: DG: disease-gene; DC: disease-chemical; GC: gene-chemical

cd src
python runProspective.py --e DG # Could be one of {'DG','DC','GC'}

Experiment parameter options: DG: disease-gene; DC: disease-chemical; GC: gene-chemical

multimodal-network-diffusion/
├── README.md                       <- This document
├── install.sh                      <- The script to set up environment and download data.
├── download_precomp.sh             <- The script to download precomputed data
├── Diffusion2018.yml               <- Conda environment file
├── pyjulia-master_20180601.zip     <- PyJulia downloaded on 2018/06/01.
│
├── src/                            <- Source code for use in this project.
│   ├── runKfoldCV.py               <- The script to run 10-fold cross-validation.
│   ├── runLeaveAModeOut.py         <- The script to run leave-one-mode-out experiments.
│   ├── runTimeStamped.py           <- The script to run time-stamped experiments.
│   ├── runProspective.py           <- The script to run prospective experiments.
│   ├── paths.py                    <- The script to load environment variables.
│   │
│   ├── install/                    <- The scripts to test installation.
│   │
│   ├── Networks/                   <- The classes for manipulating graph data.
│   │
│   ├── Algorithms/                 <- The classes of different algorithms.
│   │
│   └── Validation_Methods/         <- The classes of different validation experiments.
│
└── notebooks/                      <- Jupyter notebooks
    ├── Prospective_example.ipynb   <- Example to run a prospective experiment
    └── Prospective_example.html    <- The html of expected results of the example.

1. Mapped networks, mapping files, saved PubMed searching results, and predictions from prospective experiments (could be downloaded through install.sh):
   data.tar.gz (md5: 827e732c6480b2ffcc19ee16721869d9; 75 MB after compression)
2. Precomputed algorithm results of 10-fold cross-validation (could be downloaded through download_precomp.sh):
   10_fold.tar.gz (md5: a9b538536c0b59f3a3860ac77fffa755; 275 GB after compression)
3. Precomputed algorithm results of leave-one-mode-out experiments (could be downloaded through download_precomp.sh):
   lomo.tar.gz (md5: f996d985a4fa9c6baf02ff91f2021cf2; 9.4 GB after compression)
4. Precomputed algorithm results of time_stamped experiments (could be downloaded through download_precomp.sh):
   time_stamped.tar.gz (md5: 16a18fce049820f5cce95d7f5eba3916; 22 GB after compression)
5. Precomputed algorithm results of prospective experiments (could be downloaded through download_precomp.sh):
   prospective.tar.gz (md5: 5cbd2158c9b8895cb4aa836e1800ae48; 21.6 GB after compression)
6. Html files of prediction results of prospective experiments (could be downloaded through download_precomp.sh):
   Diffusion2018predictions_html.tar.gz (md5: 82bbe7cfd119adb9d10046b049959717; 2.2 Mb after compression)