/multimodal-network-diffusion

Repository for *Multimodal Network Diffusion Predicts Future Disease-Gene-Chemical Associations*

Primary LanguageHTML

Multimodal Network Diffusion

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).

Content

Download code

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

Installation and download network data

Requirements

  • 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

Install and download network data

cd multimodal-network-diffusion
./install.sh

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

Run tutorial

1. Open Jupyter notebook in ./notebooks/

cd notebooks/
jupyter notebook

2. Click on Prospective_example.ipynb

3. Choose the kernel of the Diffusion2018 environment

(1) Choose Python [Diffusion2018] when being asked to choose kernel

or (2) (In the menu bar of the newly opened page) Kernel -> Change kernel -> Python [Diffusion2018]

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

4. Run each cell

Q: If you don't see Python [Diffusion2018] in Jupyter Notebook

A: Enable environment for Jupyter Notebook

1. Activate environment (skip it if you already activated)

source activate Diffusion2018

2. Register the environment for Jupyter Notebook

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

Q: If you need to run Jupyter Notebook through ssh

A: Please follow the instructions here

Run experiments

0. Recommended resource

Memory >=256GB is recommended for running large networks.

1. (Optional) Download precomputed results of algorithms

./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 ?

2. Activate environment

source activate Diffusion2018

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

3. Example of running 10-fold cross-validation of 1-mode networks

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

4. Example of running leave-one-mode-out experiment of disease-gene prediction

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

5. Example of running time-stamped experiment of disease-chemical prediction

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

6. Example of running prospective experiment of disease-gene prediction

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

Project organization

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.

Links to download data

  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)