Analysis of Chromosome Conformation Capture data (Hi-C)
This document describes the output produced by the pipeline. Most of the plots are taken from the MultiQC report, which summarises results at the end of the pipeline. The directories listed below will be created in the results directory after the pipeline has finished. All paths are relative to the top-level results directory.
The pipeline is built using Nextflow and processes data using the following steps:
- From raw data to valid pairs
- Hi-C contact maps
- Downstream analysis
- MultiQC - aggregate report and quality controls, describing results of the whole pipeline
- Export - additionnal export for compatibility with downstream analysis tool and visualisation
From raw data to valid pairs
Using Hi-C data, each reads mate has to be independently aligned on the
The current workflow implements a two steps mapping strategy. First, the reads
are aligned using an end-to-end aligner.
Second, reads spanning the ligation junction are trimmmed from their 3’ end,
and aligned back on the genome.
Aligned reads for both fragment mates are then paired in a single paired-end
Singletons and low quality mapped reads are filtered (
Note that if the
--dnase mode is activated, HiC-Pro will skip the second
*bwt2pairs.bam- final BAM file with aligned paired data
--save_aligned_intermediates is specified, additional mapping file results
are available ;
*.bam- Aligned reads (R1 and R2) from end-to-end alignment
*_unmap.fastq- Unmapped reads after end-to-end alignment
*_trimmed.fastq- Trimmed reads after end-to-end alignment
*_trimmed.bam- Alignment of trimmed reads
*bwt2merged.bam- merged BAM file after the two-steps alignment
*.mapstat- mapping statistics per read mate
Usually, a high fraction of reads is expected to be aligned on the genome (80-90%). Among them, we usually observe a few percent (around 10%) of step 2 aligned reads. Those reads are chimeric fragments for which we detect a ligation junction. An abnormal level of chimeric reads can reflect a ligation issue during the library preparation. The fraction of singleton or low quality reads depends on the genome complexity and the fraction of unmapped reads. The fraction of singleton is usually close to the sum of unmapped R1 and R2 reads, as it is unlikely that both mates from the same pair were unmapped.
Valid pairs detection with HiC-Pro
Each aligned reads can be assigned to one restriction fragment according to the reference genome and the digestion protocol.
Invalid pairs are classified as follow:
- Dangling end, i.e. unligated fragments (both reads mapped on the same restriction fragment)
- Self circles, i.e. fragments ligated on themselves (both reads mapped on the same restriction fragment in inverted orientation)
- Religation, i.e. ligation of juxtaposed fragments
- Filtered pairs, i.e. any pairs that do not match the filtering criteria on inserts size, restriction fragments size
- Dumped pairs, i.e. any pairs for which we were not able to reconstruct the ligation product.
Only valid pairs involving two different restriction fragments are used to
build the contact maps.
Duplicated valid pairs associated to PCR artefacts are discarded
--keep_dup to not discard them).
In case of Hi-C protocols that do not require a restriction enzyme such as
DNase Hi-C or micro Hi-C, the assignment to a restriction is not possible
Short range interactions that are likely to be spurious ligation products
can thus be discarded using the
*.validPairs- List of valid ligation products
*.DEpairs- List of dangling-end products
*.SCPairs- List of self-circle products
*.REPairs- List of religation products
*.FiltPairs- List of filtered pairs
*RSstat- Statitics of number of read pairs falling in each category
Of note, these results are saved only if
--save_pairs_intermediates is used.
validPairs are stored using a simple tab-delimited text format ;
The ligation efficiency can be assessed using the filtering of valid and invalid pairs. As the ligation is a random process, 25% of each valid ligation class is expected. In the same way, a high level of dangling-end or self-circle read pairs is associated with a low quality experiment, and reveals a problem during the digestion, fill-in or ligation steps.
In the context of Hi-C protocol without restriction enzyme, this analysis step is skipped. The aligned pairs are therefore directly used to generate the contact maps. A filter of the short range contact (typically <1kb) is recommanded as this pairs are likely to be self ligation products.
validPairs file are generated per reads chunck (and saved only if
--save_pairs_intermediates is specified).
These files are then merged in the
allValidPairs file, and duplicates are
--keep_dups to disable duplicates filtering).
*allValidPairs- combined valid pairs from all read chunks
Additional quality controls such as fragment size distribution can be extracted from the list of valid interaction products. We usually expect to see a distribution centered around 300 bp which corresponds to the paired-end insert size commonly used. The fraction of duplicates is also presented. A high level of duplication indicates a poor molecular complexity and a potential PCR bias. Finally, an important metric is to look at the fraction of intra and inter-chromosomal interactions, as well as long range (>20kb) versus short range (<20kb) intra-chromosomal interactions.
.pairs is a standard tabular format proposed by the 4DN Consortium
for storing DNA contacts detected in a Hi-C experiment
This format is the entry point of the downstream steps of the pipeline after
detection of valid pairs.
*pairix- compressed and indexed pairs file
Various statistics files are generated all along the data processing.
All results are available in
- *mapstat - mapping statistics per read mate
- *pairstat - R1/R2 pairing statistics
- *RSstat - Statitics of number of read pairs falling in each category
- *mergestat - statistics about duplicates removal and valid pairs information
Intra and inter-chromosomal contact maps are built for all specified resolutions. The genome is split into bins of equal size. Each valid interaction is associated with the genomic bins to generate the raw maps. In addition, Hi-C data can contain several sources of biases which has to be corrected. The HiC-Pro workflow uses the ìced and Varoquaux and Servant, 2018 python package which proposes a fast implementation of the original ICE normalisation algorithm (Imakaev et al. 2012), making the assumption of equal visibility of each fragment.
Importantly, the HiC-Pro maps are generated only if the
is specified on the command line.
*.matrix- genome-wide contact maps
*_iced.matrix- genome-wide iced contact maps
The contact maps are generated for all specified resolutions
A contact map is defined by :
- A list of genomic intervals related to the specified resolution (BED format).
- A matrix, stored as standard triplet sparse format (i.e. list format).
Based on the observation that a contact map is symmetric and usually sparse, only non-zero values are stored for half of the matrix. The user can specified if the ‘upper’, ‘lower’ or ‘complete’ matrix has to be stored. The ‘asis’ option allows to store the contacts as they are observed from the valid pairs files.
This format is memory efficient, and is compatible with several software for downstream analysis.
Hi-C contact maps
Contact maps are usually stored as simple txt (
HiC-Pro), .hic (
Juicer/Juicebox) and .(m)cool (
The .cool and .hic format are compressed and indexed and usually much more efficient than the txt format.
In the current workflow, we propose to use the
cooler format as a standard to build the raw and normalised maps
after valid pairs detection as it is used by several downstream analysis and visualisation tools.
Raw contact maps are therefore in
results/contact_maps/raw which contains the different maps in
cool formats, at various resolutions.
Normalised contact maps are stored in
results/contact_maps/norm which contains the different maps in
The bin coordinates used for all resolutions are available in
txt contact maps generated with
cooler are identical to those generated by
However, differences can be observed on the normalised contact maps as the balancing algorithm is not exactly the same.
Downstream analysis are performed from
cool files at specified resolution.
The distance decay plot shows the relationship between contact frequencies and genomic distance. It gives a good indication of the compaction of the genome. According to the organism, the slope of the curve should fit the expectation of polymer physics models.
The results generated with the
HiCExplorer hicPlotDistVsCounts tool (plot and table) are available in the
Compartments calling is one of the most common analysis which aims at detecting A (open, active) / B (close, inactive) compartments. In the first studies on the subject, the compartments were called at high/medium resolution (1000000 to 250000) which is enough to call A/B compartments. Analysis at higher resolution has shown that these two main types of compartments can be further divided into compartments subtypes.
Although different methods have been proposed for compartment calling, the standard remains the eigen vector decomposition from the normalised correlation maps.
Here, we use the implementation available in the
Results are available in
results/compartments/ folder and include :
*cis.vecs.tsv: eigenvectors decomposition along the genome
*cis.lam.txt: eigenvalues associated with the eigenvectors
TADs have been described as functional units of the genome. While contacts between genes and regulatority elements can occur within a single TAD, contacts between TADs are much less frequent, mainly due to the presence of an insulation protein (such as CTCF) at their boundaries. Looking at Hi-C maps, TADs look like triangles around the diagonal. According to the contact map resolutions, TADs appear as hierarchical structures with a median size around 1Mb (in mammals), as well as smaller structures usually called sub-TADs of smaller size.
TADs calling remains a challenging task, and even if many methods have been proposed in the last decade, little overlap has been found between their results.
Currently, the pipeline proposes two approaches :
- Insulation score using the
cooltoolspackage. Results are availabe in
HiCExplorer TADs calling. Results are available at
Usually, TADs results are presented as simple BED files, or bigWig files, with the position of boundaries along the genome.
multiqc_report.html: a standalone HTML file that can be viewed in your web browser.
multiqc_data/: directory containing parsed statistics from the different tools used in the pipeline.
multiqc_plots/: directory containing static images from the report in various formats.
MultiQC is a visualisation tool that generates a single HTML report summarising all samples in your project. Most of the pipeline QC results are visualised in the report and further statistics are available in the report data directory.
Results generated by MultiQC collate pipeline QC from supported tools e.g. FastQC. The pipeline has special steps which also allow the software versions to be reported in the MultiQC output for future traceability. For more information about how to use MultiQC reports, see http://multiqc.info.
- Reports generated by Nextflow:
- Reports generated by the pipeline:
pipeline_report*files will only be present if the
--email_on_failparameter’s are used when running the pipeline.
- Reformatted samplesheet files used as input to the pipeline:
- Reports generated by Nextflow:
Nextflow provides excellent functionality for generating various reports relevant to the running and execution of the pipeline. This will allow you to troubleshoot errors with the running of the pipeline, and also provide you with other information such as launch commands, run times and resource usage.