scKTLD is a method designed for the identification of TAD-like domains on single-cell Hi-C data. It treats the Hi-C contact matrix as a graph, embeds its structures into a low-dimensional space by combining sparse matrix factorization and spectral propagation, and identifies the TAD-like domains in the embedding space via a kernel model optimized by PELT
2.1 OS
2.2 Required Python Packages Make sure that all the packages listed in the requirements.txt are installed.
2.3 Install from Github
(1) Download the folder scKTLD by git clone
$ git clone https://github.com/lhqxinghun/scKTLD/
(2) Install the package scKTLD with the following command:
$ conda create -n scKTLD python=3.6
$ conda activate scKTLD
$ pip install Cython
$ cd scKTLD
$ pip install . #or you can try python setup.py install
2.4 Or install from the standard package source PyPI
$ conda create -n scKTLD python=3.6
$ conda activate scKTLD
$ pip install Cython scKTLD
2.5 Run example
$ cd scKTLD
$ python example.py
# If it works properly, You can find the result files in the output directory, including the .txt file that contains the
identified TAD-like domain boundaries and the .tiff file for visualization. The .tiff file in the example is shown as follows:
More detailed examples can be find in the jupyter notebook example.ipynb
(1) The key function in this package is callTLD, it has the following input and output:
If brecon is false, the callTLD function will only return a list of domain boundaires, else it will return the domain boundaries as well as a reconstructed Hi-C map
(2) For sparse format of a contact matrix, scKTLD provides function edge2adj to convert it to an adjacency matrix (dense format), which can be directly input to the fucntion callTLD
The contact matrix in dense format
hongqianglv@mail.xjtu.edu.cn OR liuerhu@stu.xjtu.edu.cn