Skip to content

DiffDomain is a statistically sound method for detecting differential TADs between conditions

License

Notifications You must be signed in to change notification settings

Tian-Dechao/diffDomain

Repository files navigation

diffDomain

A short description

diffDomain is a new computational method for identifying reorganized TADs using chromatin contact maps from two biological conditions.

A long description diffDomain

The workflow of diffDomain is illustrated down below.

The goal is to test if a TAD identified in one biological condition has structural changes in another biological condition.

The core of diffDomain is formulating the problem as a hypothesis testing problem where the null hypothesis is that the TAD doesn't undergo significant structural reorganization at later condition. The input are Hi-C contact matrices of the TAD region in the two biological conditions (A). The Hi-C contact matrices are log-transformed to adjust for the exponential decay of Hi-C contacts between chromosome bins with increased distances.

Their entry-wise difference is calculated (B).

The difference matrix D is normalized by iteratively standardizing its k-off diagonal parts, -N+2 <= k <= N-2, adjusting absolute differences in contact frequencies due to different sequencing depths in the two biological conditions (C).

Note that, standardization is TAD-specific. Each TAD has its own parameters that are only estimated from its contact matrices in a pair of biological conditions.

Intuitively, if a TAD is not significantly reorganized, normalized D would resemble a random matrix with white noise entries, enabling us to borrow theoretical results in random matrix theory. Indeed, normalized D is a generalized Wigner matrix (D), a well studied high-dimensional random matrices.

Its largest singular value is proved to be fluctuating around 2 under the null hypothesis. Armed with the fact, diffDomain reformulates the reorganized TAD identification problem into a hypothesis testing problem:

  1. H0: the largest singular value equals to 2;
  2. H1: the largest singular value is greater than 2.

For a user given set of TADs, P values are adjusted for multiple comparisons using BH method as default.
Once we identify the subset of reorganized TADs, we classify them into six subtypes to aid biological analysis and interpretations (F).
A few examples of reorganized TADs identified by diffDomain in two datasets are shown in (G).

workflow

Installation instructions

diffDomain is tested on MacOS & Linux (Centos).

Dependences

diffDomain-py2 is dependent on

  • Python 2.7
  • hic-straw==0.0.6

diffDomain-py3 is dependent on

  • Python 3
  • hic-straw==1.3.1

and

  • cooler
  • hicexplorer
  • TracyWidom
  • pandas
  • numpy
  • docopt
  • tqdm
  • matplotlib
  • statsmodels
  • h5py
  • seaborn

Installation

First of all, we recommend you to have a package manager, such as conda, and create a new independent environment for diffDomain.

Method1: to install the conda environment

Step1:

git clone https://github.com/Tian-Dechao/diffDomain
cd diffDomain

Step2:

For Linux

conda env create --name diffdomain -f environment_linux.yml

For MacOS

conda env create --name diffdomain -f environment_macos.yml

Step3:

conda activate diffdomain

In this environment, all the need of diffDomain(Python3 version) have been installed.

Method2: to install python3 version from Pypi

pip install diffDomain-py3

Note: If you encounter errors when installing hicstraw that diffDomain relies on, you can use conda to install it:

conda install -c bioconda hic-straw

Method3: Docker image named guming5/diffdomain-centos7:v1

docker pull guming5/diffdomain-centos7:v1
docker run -it guming5/diffdomain-centos7:v1 /bin/bash
# shift to the normal user named work
su work
cd ~
source activate diffdomain

In this image, there is a contact conda environment named diffdomain (/home/work/.conda/envs/diffdomain) meeting all requests, in which you can use the diffDomain Python3 version directly.

Documentation

Please see the wiki for extensive documentation and example tutorials.

Contact information

More information please contact Dunming Hua at [email protected], Ming Gu at [email protected] or Dechao Tian at [email protected].

References

DOI