Sub-compartment Identifier (SCI)

Authors: Haitham Ashoor, Sheng Li Contact: haitham.ashoor@jax.org, sheng.li@jax.org

Description

SCI is a program to identify sub-compartments from HiC data. SCI utilizes graph embedding followed by K-means clustering in order to predict sub-compartments from HiC data.

Dependencies

python 2.7

Python Libraries

scikit-learn >=0.19.0
Numpy >= 1.15
tqdm>=4.24

C++ libraries

Installation

$ python setup.py install

Input format

SCI accepts bedpe-like format

chr1: is the chromosome name for the first interacting HiC bin
start1: is the starting coordinate for the first interacting HiC bin
end1: is the ending coordinate for the first interacting HiC bin
chr2: is the chromosome name for the second interacting HiC bin
start2: is the starting coordinate for the second interacting HiC bin
end2: is the ending coordinate for the second interacting HiC bin
HiC count: number of HiC reads for the interacting HiC bins. SCI does not perform HiC normalization, if user wants to use normalized HiC data, HiC count should corresponds to the normalized HiC read-count.

SCI provides a script to convert .hic format into SCI accepted format under scripts/hic2sci.sh. In order to convert .hic file into please follow the following instructions:

export installed juicer-tools into JUICERTOOLS environment variable

$ export JUICERTOOLS=/path/to/juicer-tools

Then, run hic2sci script to get SCI formatted input data:

$ scripts/hic2sci.sh <input .hic file> <output file> <resolution>

Docker Container

Different operating systems my require certain adjustment for the script, thus we build a docker container to solve this problem. The packages are already installed thus we can directly run sci using the following commands.

The command to start docker container is:

docker run -it -p 8080:8080 -v <directory of the Rao_2014.hic data file>:/data yuz12012/sci_container:latest

For the container, please run

export LD_LIBRARY_PATH=/sci/gsl/lib
export CPPFLAGS="-I/usr/local/zlib/1.2.8-4/include"
export JUICERTOOLS=/sci/juicer_tools_1.22.01.jar
sh scripts/hic2sci.sh /data/Rao_2014.hic sci_input 100000

bash
conda activate sci

For people using singularity to load the docker container, please change the output directory with writable authority.

Test run

To preform test run for SCI please follow the following steps: The sample input sample is at: ftp://ftp.jax.org/zhaoyu/demo_data.txt.zip

Please run the following command in sci root directory.

$ python -m sci.sci -n test -f /sci-data/demo_data.txt -r 100000 -g chromosome_sizes/hg19.chrom.sizes -o both -s 1 -k 5

Parameters description:

Parameter	Mandatory/Optional	Description
-n, --name	yes	Name of the experiment, it will be used as a prefix for all output files
-r, --resolution	yes	Required resolution to predict compartments,provided bins' size should have resolution greater than or equal the provided value
-g, --genome_size	yes	File containing chromosome sizes of the target genome
-o, --order	No. Default: 1	Graph order to consider when performing graph embedding. Available options are 1,2 or both
-s, --samples	No. Default: 25	Number of edges to sample in millions order from the graph
-k, --clusters	No. Default: 2	Nubmer of sub-compartments to be predicted

Output

SCI output sub-compartments annotation into BED format with the following fields:

chr: chromosome for sub-compartment annotaiton
start: genomic location where sub-compartment bin starts
end: genomic location where sub-compartment bin ends
label: sub-compartment unique label. Bins that do not have sub-compartment label due to low mapability are labeled with NA.

LabShengLi / sci