View on GitHub

cLoops

Accurate and flexible loops calling tool for 3D genomic data.

[![Language](https://img.shields.io/github/languages/top/YaqiangCao/cLoops)](https://img.shields.io/github/languages/top/YaqiangCao/cLoops) [![Stars](https://img.shields.io/github/stars/YaqiangCao/cLoops?logo=GitHub&color=yellow)](https://github.com/YaqiangCao/cLoops/stargazers) [![LOC](https://tokei.rs/b1/github/YaqiangCao/cLoops?category=code)](https://github.com/Aaronepower/tokei) ViewCount [![GitHub Clones](https://img.shields.io/badge/dynamic/json?color=success&label=Clone&query=count&url=https://gist.githubusercontent.com/YaqiangCao/35ae75af5d9b37d9159a108dabfef956/raw/clone.json&logo=github)](https://github.com/MShawon/github-clone-count-badge)

cLoops: loop-calling for ChIA-PET, Hi-C, HiChIP and Trac-looping

Introduction

Chromosome conformation capture (3C) derived high-throughput sequencing methods such as ChIA-PET,HiChIP and Hi-C provide genome-wide view of chromatin organization. Fine scale loops formed by interactions of regulatory elements spanning hundreds kilobases can be detected from these data. Here we introduce cLoops (‘see loops’),a common loops calling tool for ChIA-PET, HiChIP and high-resolution Hi-C data. Paired-end tags (PETs) are first classified as self-ligation and inter-ligation clusters using an optimized unsupervisied clustering algorithm. The significances of the inter-ligation clusters are then estimated using permutated local background.

If you find cLoops useful, please give us a star at github and cite our paper :

Official version: Yaqiang Cao, Zhaoxiong Chen, Xingwei Chen, Daosheng Ai, Guoyu Chen, Joseph McDermott, Yi Huang, Guo Xiaoxiao, Jing-Dong J Han, Accurate loop calling for 3D genomic data with cLoops, Bioinformatics, , btz651, https://doi.org/10.1093/bioinformatics/btz651

Preprint bioRxiv: Yaqiang Cao, Xingwei Chen, Daosheng Ai, Zhaoxiong Chen, Guoyu Chen, Joseph McDermott, Yi Huang, Jing-Dong J. Han (2018) “Accurate loop calling for 3D genomic data with cLoops” bioRxiv 465849; doi: https://doi.org/10.1101/465849

You can also find the cLoops wiki in Chinese here

Please kindly refer to cLoops2 for more analytical modules.


Install

If you are familar with conda, cLoops could be installed very easily with following after clone and cd in it.

git clone https://github.com/YaqiangCao/cLoops
cd cLoops
conda env create --name cLoops --file cLoops_env.yaml
conda activate cLoops 
python setup.py install

Then every time just use conda activate cLoops to run cLoops enviroment.

Or you prefer the old school, install from scratch. scipy,numpy, seaborn, pandas and joblib are required. Joblib version 0.11 is requried to avoid parallel computating bugs caused by it for newer version. Install it through pip2.7 install –user joblib==0.11. If you have problems for installing scipy, please refer to Anaconda or SAGE.

wget https://github.com/YaqiangCao/cLoops/archive/0.93.tar.gz
tar xvzf 0.93.tar.gz
cd cLoops-0.93
python setup.py install    

To test whether cLoops is successfully installed:

cd examples
sh run.sh

Please refer to here to install cLoops to customized path.


Usage

Run cLoops -h to see all options. Key parameters are eps and minPts . minPts defines at least how many PETs are required for a candidate loop, eps defines the distance requried for two PETs being neighbors. For practically usage to tune parameters, using the PETs in the smallest chromosome except chrY and chrM, then run a series of eps and minPts,all rounds clustering result will be combined to determine your parameters.

Since version 0.8, cLoops added a parameter –mode(-m), which is the pre-set parameters for different types of data. -m 0 accepts user settings; -m 1 equals -eps 500,1000,2000 -minPts 5 for sharp peak like ChIA-PET data; -m 2 equals -eps 1000,2000,5000 -minPts 5 for broad peak like ChIA-PET data; -m 3 equals -eps 5000,7500,10000 -minPts 20,30,40,50 -hic for deep sequenced Hi-C data (~200 million cis PETs); -m 4 equals -eps 2500,5000,7500,10000 -minPts 20,30 -hic for ~100 million cis PETs HiChIP data;for ~30-40 miilion cis PETs HiChIP data, we suggested -eps 2500,5000,7500,10000 -minPts 10,15,20 -hic. You can always add more eps and smaller minPts to get more candidate loops and maybe more significant loops, however, it takes longer time.


Input

Mapped PETs in BEDPE format, compressed files with gzip are also accepected, following columns are necessary: chrom1 (1st),start1 (2),end1 (3),chrom2 (4),start2 (5),end2 (6),strand1 (9),strand2 (10). For the column of name or score, “.” is accepcted. Columns are seperated by “\t”. For example as following :

chr1	9945	10095	chr1	248946216	248946366	.	.	+	+
chr1	10034	10184	chr1	180987	181137	.	.	+	-
chr1	10286	10436	chr1	181103	181253	.	.	+	-

Output

The main output is a loop file and a PDF file or PDFs for the plot of self-ligation and inter-ligation PETs distance distributions. For the .loop file, columns and explaination are as follwing:

column name explaination
0th loopId Id for a loop, like chr1-chr1-1
1th ES Enrichment score for the loop, caculated by observed PETs number divided by the mean PETs number of nearby permutated regions
2th FDR false discovery rate for the loop, caculated as the number of permutated regions that there are more observed PETs than the region
3th binomal_p-value binomal test p-value for the loop
4th distance distance (bp) between the centers of the anchors for the loop
5th hypergeometric_p-value hypergeometric test p-value for the loop
6th iva genomic coordinates for the left anchor, for example, chr13:50943050-50973634
7th ivb genomic coordinates for the right anchor
8th poisson_p-value poisson test p-value for the loop
9th ra observed PETs number for the left anchor
10th rab observed PETs number linking the left and right anchors
11th rb observed PETs number for the right anchor
12th poisson_p-value_corrected Bonferroni corrected poisson p-value according to number of loops for each chromosome
13th binomal_p-value_corrected Bonferroni corrected binomal p-value according to number of loops for each chromosome
14th hypergeometric_p-value_corrected Bonferroni corrected hypergeometric p-value according to number of loops for each chromosome
15th significant 1 or 0, 1 means we think the loop is significant compared to permutated regions. You can ignore this and customize your cutoffs using above values by visualization a small chromosome in the Juicebox or washU.

Examples

All following examples source data, result and log file can be found in the examples.

1. ChIA-PET data

We provide a test data from GM12878 CTCF ChIA-PET (GSM1872886), just the chromosome 21 mapped to hg38. Run the command as following then you will get the result if cLoops is successfuly installed. The eps is auto estimated and default minPts is 5,-w option will generate loops for visualization in washU browser,-j option will generate loops for visualization in Juicebox .

wget https://github.com/YaqiangCao/cLoops/blob/master/examples/GSM1872886_GM12878_CTCF_ChIA-PET_chr21_hg38.bedpe.gz
cLoops -f GSM1872886_GM12878_CTCF_ChIA-PET_chr21_hg38.bedpe.gz -o chiapet -w -j -s -m 1 -plot

For ChIA-PET data with sharp peak, like the CTCF here, you will get the inter-ligation and self-ligation PETs distance distribution like following, the two kinds of PETs well seperated using auto estimated eps:

If your experimental data doesn’t look like this by auto estimated eps, which could be true for some ChIA-PET data with broad peak (like H3K27ac), please use the small chromosome (chr21 in human and chr19 in mouse) run a series of eps, then chose the smallest one that generate the well seperated distance distribution to run cLoops, or just using the series.

We recommend washU to visualize the loops, by the script jd2washU we can convert the cLoops temp files to washU long range track, and bedtools,bgzip & tabix are needed in the command enviroment.

jd2washU -d chiapet -o chiapet       

With other ChIP-seq data, you can get following plot:

2. HiChIP data

We provide test data of GM12878 cohesin HiChIP two biological replicates, just the chromosome 21 mapped to hg38. Run the command as following to call merged loops. -s option is used to keep working directory and temp files, which could be used by scripts of deLoops, jd2washU (BEDTOOLS needed), jd2juice (Juicer needed), jd2fingerprint and jd2saturation. -hic option means using cutoffs design for Hi-C like data, see above.

wget https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSE80820_GM12878_cohesin_HiChIP_chr21_hg38_bio1.bedpe.gz 
wget https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSE80820_GM12878_cohesin_HiChIP_chr21_hg38_bio2.bedpe.gz 
cLoops -f GSE80820_GM12878_cohesin_HiChIP_chr21_hg38_bio1.bedpe.gz,GSE80820_GM12878_cohesin_HiChIP_chr21_hg38_bio2.bedpe.gz -o hichip -m 4 -j -s -w 

To convert cLoops temp files to hic file for juicebox, juicer tools are required. Java 1.7 or 1.8 is also required to run juicer tools. Script named juicer_tools with the following content should be put inside a directory included in command environment, then enable its executable privilege.

#!/bin/sh
java -jar /PATH/TO/JUICER_TOOLS_JAR_DIRECTORY/juicer_tools.1.8.9_jcuda.0.8.jar $@

Then use jd2juice:

jd2juice -d hichip -o hichip -org hg38 

With the adjustment of resolution, color range and how to show the loops, then you can get following visualization:

3. Hi-C data

We provide test data from GM12878 Hi-C, just the chromosome 21 mapped to hg38. Run the the command as following to call loops.

wget https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSM1551552_GM12878_HiC_chr21_hg38.bedpe.gz 
cLoops -f GSM1551552_GM12878_HiC_chr21_hg38.bedpe.gz -o hic -w -j -eps 5000,7500,10000 -minPts 20,30 -s -hic 

or just run following for version >= 0.9:

cLoops -f GSM1551552_GM12878_HiC_chr21_hg38.bedpe.gz -o hic -w -j -s -m 3

4. Fingerprint plot for data qualities comparasion of loops calling

Run following and you will get a PDF plot, the far from the random line, the better for the data used to call loops by cLoops. You can using this to estimate data qualities between samples.

jd2fingerprint -d chiapet,hichip,hic -plot 1 -o compare -bs 2000

5. call stripes

Since v0.91 (2018-05-17 release), we introduce a new script callStripe, which can identify stripes (a structure defined in The Energetics and Physiological Impact of Cohesin Extrusion). However, the original paper hasn’t released their data, so we demonstrate the the result using H3K27ac HiChIP data in K562, which from the heatmap we can observe a lot of similar stripes. We provided the H3K27ac HiChIP data in K562 chr21 for testing. Parameters tuning maybe needed for other data, please email caoyaqiang0410@gmail.com for tuning parameters for your data.

wget https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSE101498_K562_HiChIP_H3K27ac_rep1.bedpe.gz
wget https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSE101498_K562_HiChIP_H3K27ac_rep2.bedpe.gz
wget https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSE101498_K562_HiChIP_H3K27ac_rep3.bedpe.gz
#first call loops to save the middle files, you can kill cLoops once the .jd files are generated
cLoops -f GSE101498_K562_HiChIP_H3K27ac_rep1.bedpe.gz,GSE101498_K562_HiChIP_H3K27ac_rep2.bedpe.gz,GSE101498_K562_HiChIP_H3K27ac_rep3.bedpe.gz -o K562_HiChIP_H3K27ac_chr21 -minPts 20,30 -eps 2500,5000,7500,10000 -hic -s -j -c chr21
#call stripes
callStripes -d K562_HiChIP_H3K27ac_chr21 -o K562_HiChIP_H3K27ac_chr21 -c chr21 -j
#for visualization in juicebox
jd2juice -d K562_HiChIP_H3K27ac_chr21/ -o K562_HiChIP_H3K27ac_chr21 -org hg38

After above command, you will get two files (with suffix juicebox.txt,K562_HiChIP_H3K27ac_chr21_x_horizontal_juicebox.txt and K562_HiChIP_H3K27ac_chr21_y_vertical_juicebox.txt that could be used for furthur analysis and loaded in Juicebox as 2D annotation as following example: Two extra files with file type as .stripe (K562_HiChIP_H3K27ac_chr21_x_horizontal.stripe and K562_HiChIP_H3K27ac_chr21_y_vertical.stripe) is similar to that of .loop file. Please note, it’s a initial experimental function added in v0.91, not well tested for all data. We’ll make improvements when the deep-sequenced Hi-C data is available.

6. Trac-looping data

We provide test data of Trac-looping data from resting CD4+ cell, just the chromosome 21 mapped to hg19 obtained from GEO:GSE87254. Run the the command as following to call loops. This new datasets also show cLoops is applied to new developed 3D mapping data.The option “-max_cut” is a new option in v0.92 to select more distant loops from the distance cutoffs determined from multiple eps and minPts combinations.

#download data from our site
wget -c https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSM2326178_CD4_Resting_Trac-looping_rep1-tech1_chr21_hg19.bedpe.gz
wget -c https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSM2326179_CD4_Resting_Trac-looping_rep2_chr21_hg19.bedpe.gz
wget -c https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSM2326180_CD4_Resting_Trac-looping_rep1-tech2_chr21_hg19.bedpe.gz 
wget -c https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSM2782295_CD4_Resting_Trac-looping_rep3_chr21_hg19.bedpe.gz 
wget -c https://github.com/YaqiangCao/cLoops_supplementaryData/blob/master/examples/GSM2782296_CD4_Resting_Trac-looping_rep4_chr21_hg19.bedpe.gz 
#run cLoops, -cut 2000 was used to remove close PETs before calling loops
cLoops -f GSM2326178_CD4_Resting_Trac-looping_rep1-tech1_chr21_hg19.bedpe.gz,GSM2326179_CD4_Resting_Trac-looping_rep2_chr21_hg19.bedpe.gz,GSM2326180_CD4_Resting_Trac-looping_rep1-tech2_chr21_hg19.bedpe.gz,GSM2782295_CD4_Resting_Trac-looping_rep3_chr21_hg19.bedpe.gz,GSM2782296_CD4_Resting_Trac-looping_rep4_chr21_hg19.bedpe.gz -o Trac-looping_chr21 -eps 500,1000,2000,5000 -minPts 5 -p 1 -w -j -cut 2000 -s -c chr21 -max_cut
#conver cLoops tmp files to washU interaction tracks for visualizaiton
jd2washU -d Trac-looping_chr21 -o Trac-looping_chr21 

By uploading the interaction tracks and cLoops called loops (with suffix of _loops_washU.txt ) to washU, you can visualize the result for example as following. Meanwhile, cLoops called loops are more distant than their original called loops. We’re keeping improving cLoops.


Other data

In theory cLoops could be applied to more 3D genomic data as long as there are enriched clusters in the heatmap, however, parameters and significance cutoffs should be tuned. We’re now trying to make cLoops work for GRID-seq and Capture HiC. If you have designed a new sequencing based 3D genomic method and want to try cLoops, please contact caoyaqiang0410@gmail.com first.


Questions & Answers

Please address questions and bugs to Yaqiang Cao (caoyaqiang0410@gmail.com) or Zhaoxiong Chen (chenzhaoxiong@picb.ac.cn), using the subject as “cLoops: questions about ###” to escape misjudged as spams.

Following are selected questions:


  1. HiC-Pro to bedpe
    The allValidPairs can be converted to BEDPE file. You can define a extension size (like half of the reads length) along the reads strand direction. In cLoops’ first step, all coordinates are converted from (startA+endA)/2,(startB+endB)/2 to (x,y), so actually the extension size doesn’t matter. Since v0.92 (2018-11-16 updated), we provide a scirpt named hicpropairs2bedpe to convert HiC-Pro output to cLoops input BEDPE file. Once cLoops installed, hicpropairs2bedpe is available through command line.

  2. inter-chromosomal loops
    So far cLoops doesn’t support calling inter-chromosomal loops, as there are few significant inter-chromosomal loops called for our tested data and it takes a long time to run. However, we’ll try to implement a script for calling this kind of loops for next version as soon as there’s available testing data.

  3. For multiple eps and minPts parameters, how cLoops determine the output loops?
    For example, eps=5000,7500,10000 and minPts=5,10,20, so total there will be 9 clustering carried out to find potential loop regions, candidate with PETs less than max(minPts) (here 20) will be filtered. Further, for overlapped loops, after the significance test, cLoops will output the one with highest significance by binomial test. The reason for multiple times of clustering with eps and minPts is that,1) though DBSCAN clustering (or other) is great, there will be some random result due to the visit order of points, even though we try to control it; 2) no idea theory estimation of the parameters. 3) Hi-C data quality may different from one set to another.


selected cLoops citations

  1. Cao, Yaqiang, et al. “Widespread roles of enhancer-like transposable elements in cell identity and long-range genomic interactions.” Genome research 29.1 (2019): 40-52.
  2. Alavattam, Kris G., et al. “Attenuated chromatin compartmentalization in meiosis and its maturation in sperm development.” Nature structural & molecular biology 26.3 (2019): 175-184.
  3. Luo, Zhengyu, et al. “Reorganized 3D genome structures support transcriptional regulation in mouse spermatogenesis.” Iscience (2020): 101034.
  4. Johnstone, Sarah E., et al. “Large-Scale Topological Changes Restrain Malignant Progression in Colorectal Cancer.” Cell (2020).
  5. Zhen, Tao, et al. “RUNX1 and CBFβ-SMMHC transactivate target genes together in abnormal myeloid progenitors for leukemia development.” Blood, The Journal of the American Society of Hematology 136.21 (2020): 2373-2385.
  6. Ma, Sai, et al. “Chromatin Potential Identified by Shared Single-Cell Profiling of RNA and Chromatin.” Cell 183.4 (2020): 1103-1116.

cLoops updates

Please check at the released versions.