Sampling Hi-C Contact Matrices for Aparametric Normalization

Generate expected Hi-C (genome-wide chromosome conformation capture) data maintaining marginal coverage, and the decay in contact probability with increasing genomic distance. Expected Hi-C data is used for normalization and graphical presentation of statistical significance of observed Hi-C contact enrichments. The method is described in Mendelson Cohen et al. (2017) .

The shaman package - sampling HiC contact matrices for a-parametric normalization

Shaman package implements functions for resampling Hi-C matrices in order to generate expected contact distributions given constraints on marginal coverage and contact-distance probability distributions. The package also provides support for visualizing normalized matrices and statistical analysis of contact distributions around selected landmarks. It is an R package embedding algorithms implemented in C++, which is built over the Tanay lab's misha genomic database system (the misha package).

The normalization workflow consists of the following steps:

Import observed HiC data to misha db. In order to utilize full functionality of the shaman package, you need to construct a Misha database for your genome of interest, and import your Hi-C data to Misha binary Quad-tree format.
Shuffle observed data to generate expected data. Note that shaman can also run distributed using Sun Grid Engine.
Compute the normalized score of each observed data point. This computation can be conducted inline for a specific genomic region, or pre-processed on the entire dataset using distributed computation.
Visualize normalized contact maps. Precomputed normalized scores allow for flexible analysis of contact distributions.
Quantify contact enrichment according to genomic features.

The package includes a small example misha database (shaman_get_test_track_db()) with the observed, expected and score tracks of chr2:176.5e06-177e06, around the hoxd locus, from the ELA K562 dataset. The full example database (shaman_get_test_track_db(full = TRUE), downloaded from the lab's public S3 bucket on first use, about 100MB) is a low-footprint matrix to exemplify the shaman workflow: 4.6 million contacts from the ELA K562 dataset covering the hoxd locus (chr2:175e06-178e06) and convergent CTCF regions. Processing the complete matrix from this study requires downloading the full contact list and regenerating the reshuffled matrix.

Code

Source code can be found at: https://github.com/tanaylab/shaman

Requirements

System: multi-core unix / linux based system or SGE (sun grid engine) cluster for distributed computing are required for large HiC datasets.

Installation

You can install the released version of shaman from CRAN with:

install.packages("shaman")

And the development version from GitHub with:

remotes::install_github("tanaylab/shaman")

Both install misha automatically.

shaman_plot_tracks_and_annotations() and shaman_plot_map_score_with_annotations() also need the Bioconductor package Gviz, which is not installed by default:

if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install("Gviz")

Using the package

Please refer to https://tanaylab.github.io/shaman/articles/shaman-package.html for usage and workflow.

Citation

If you use shaman, please cite:

Mendelson Cohen N, Olivares-Chauvet P, Lubling Y, Baran Y, Lifshitz A, Hoichman M, Tanay A (2017). SHAMAN: bin-free randomization, normalization and screening of Hi-C matrices. bioRxiv. doi:10.1101/187203

citation("shaman") gives the same reference, including a BibTeX entry.