ddn3 1.0.2

Dependency network analysis under two conditions

DDN 3.0

We developed an efficient and accurate differential network analysis tool – Differential Dependency Networks (DDN). DDN is capable of jointly learning sparse common and rewired network structures, which is especially useful for genomics, proteomics, and other biomedical studies. DDN 3.0 significantly improves the speed of previous versions of DDN and is available as a Python package. This repository provides the source code and examples of using DDN.

Installation

Option 1: install into a new Conda environment using pip

One way is to install DDN into a new Conda environment. To create and activate an environment named ddn, run this:

conda create -n ddn python=3.11
conda activate ddn

Python 3.12 may have some issues with Numba in Windows.

DDN 3.0 can then be installed with the following command.

pip install ddn3

Option 2: install into an existing Conda environment

If you want to install DDN into an existing Conda environment, it is suggested to install dependencies from Conda first.

First, we need to install some common dependencies.

$ conda install -c conda-forge numpy scipy numba networkx matplotlib jupyter scipy pandas scikit-learn

Then run

pip install ddn3

Alternatively, you can clone the repository (https://github.com/cbil-vt/DDN3), or just download and unzip it. Then we can install DDN 3.0.

$ pip install ./

Or you may want to install it in development mode.

$ pip install -e ./

Usage

This toy example generates two random datasets and estimates two networks, one for each dataset.

import numpy as np
from ddn import ddn
dat1 = np.random.randn(1000, 10)
dat2 = np.random.randn(1000, 10)
networks = ddn.ddn(dat1, dat2, lambda1=0.3, lambda2=0.1)

For more details and examples, check the documentation, which includes three tutorials and the API reference. The tutorials can also be found in the docs/notebooks folder.

Tests

To run tests, go to the folder of DDN3 source code, then run pytest.

pytest iddn_data

It will compare the output of DDN with reference values. It tests DDN with various acceleration strategies.

Experiments

The simulation-related code can be found at https://github.com/cbil-vt/DDN3_simulation. The results of the simulation and the figures are archived here: https://zenodo.org/records/10543473.

Contributing

Please report bugs in the issues. You may also email us directly: Yizhi Wang ([e-mail]), Yingzhou Lu ([e-mail]), or Yue Wang ([e-mail]). If you are interested in adding features or fixing bugs, feel free to contact us.

License

The ddn3 package is licensed under the terms of the MIT license.

Citations

[1] DDN3.0: Determining significant rewiring of biological network structure with differential dependency networks.