RaichuNorm 1.1

Raichu

Accurately detecting enhancer-promoter loops from genome-wide interaction data, such as Hi-C, is crucial for understanding gene regulation. Current normalization methods, such as Iterative Correction and Eigenvector decomposition (ICE), are commonly used to remove biases in Hi-C data prior to chromatin loop detection. However, while structural or CTCF-associated loop signals are retained, enhancer-promoter interaction signals are often greatly diminished after ICE normalization and similar methods, making these regulatory loops harder to detect. To address this limitation, we developed Raichu, a novel method for normalizing chromatin contact data. Raichu identifies nearly twice as many chromatin loops as ICE, recovering almost all loops detected by ICE and revealing thousands of additional enhancer-promoter loops missed by ICE. With its enhanced sensitivity for regulatory loops, Raichu detects more biologically meaningful differential loops between conditions in the same cell type. Furthermore, Raichu performs consistently across different sequencing depths and platforms, including Hi-C, HiChIP, and single-cell Hi-C, making it a versatile tool for uncovering new insights into three-dimensional (3D) genomic organization and transcriptional regulation.

Installation

Raichu and all the dependencies can be installed through either mamba or pip:

$ conda config --append channels defaults
$ conda config --append channels bioconda
$ conda config --append channels conda-forge
$ mamba create -n 3Dnorm numba joblib "cooler==0.9.3" "scipy>=1.10"
$ mamba activate 3Dnorm
$ pip install RaichuNorm

Raichu is a command-line tool, and after successful installation, help information can be accessed by running raichu -h in a terminal.

Usage

Raichu is built on the cooler Python package for reading and processing contact matrices. To demonstrate how to normalize a contact matrix in .cool format, let’s download the file “GM12878.Hi-C.10kb.cool” from this link. This file contains contact matrices at 10kb resolution, generated from an in situ Hi-C dataset in the GM12878 cell line.

Now all that is needed is to execute the commands below in a terminal:

$ raichu --cool-uri GM12878.Hi-C.10kb.cool --window-size 200 -p 8 -n obj_weight -f

Here:

1. The --cool-uri parameter specifies the URI of contact matrices at a specific resolution. For a single-resolution cooler file (typically suffixed with .cool), the value should be the file path. For a multi-resolution cooler file (typically suffixed with .mcool), the value should include the file path followed by :: and the internal group path to the root of a data collection. For example: test.mcool::resolutions/10000 or test.mcool::resolutions/5000.

2. The --window-size parameter specifies the size of the sliding window. In most cases, the default value of 200 is sufficient. Increasing the window size may improve the accuracy of bias vector calculations but will also increase the runtime.

3. The -p or --nproc parameter specifies the number of processes to allocate for the calculation. Raichu uses this parameter to perform calculations for chromosomes in parallel. However, setting this parameter to a value greater than the number of chromosomes will not result in additional speed improvements.

4. The -n or --name parameter specifies the name of the column where the calculated bias vectors will be written.

5. If the -f or --force parameter is specified, the target column in the bin table will be overwritten if it already exists.

Downstream Analysis with Raichu-Normalized Matrices

Raichu stores the calculated bias vectors in the same format as cooler balance (an implementation of the ICE algorithm), ensuring seamless compability with downstream tools for analyzing compartments, TADs, and loops.

For instance, to compute chromatin compartment values based on Raichu-normalized signals, we can use the cooltools eigs-cis command and specify the --clr-weight-name parameter as “obj_weight” (matching the -n parameter setting we used when running Raichu). The full command would look like this:

$ cooltools eigs-cis --phasing-track hg38-gene-density-100K.bedGraph --clr-weight-name obj_weight -o GM_raichu GM12878-MboI-allReps-hg38.mcool::resolutions/100000

Similarly, we can use the following command to compute insulation scores with Raichu-normalized signals:

$ cooltools insulation --ignore-diags 1 -p 8 -o GM_raichu.IS.25kb.tsv --clr-weight-name obj_weight GM12878-MboI-allReps-hg38.mcool::resolutions/25000 1000000

For loop detection, we have tested the pyHICCUPS, Mustache, and Peakachu software.

Here is an example command for using pyHICCUPS (v0.3.8):

$ pyHICCUPS -p GM12878.Hi-C.5kb.cool -O GM12878_pyHICCUPS.5kb.bedpe --pw 1 2 4 --ww 3 5 7 --only-anchors --nproc 8 --clr-weight-name obj_weight --maxapart 4000000
$ pyHICCUPS -p GM12878.Hi-C.10kb.cool -O GM12878_pyHICCUPS.10kb.bedpe --pw 1 2 4 --ww 3 5 7 --only-anchors --nproc 8 --clr-weight-name obj_weight --maxapart 4000000
$ combine-resolutions -O GM12878_pyHICCUPS.bedpe -p GM12878_pyHICCUPS.5kb.bedpe GM12878_pyHICCUPS.10kb.bedpe -R 5000 10000 -G 10000 -M 100000 --max-res 10000

And here is an example command for using Mustache (v1.3.2):

$ mustache -f GM12878-MboI-allReps-hg38.mcool -r 10000 -pt 0.05 -norm obj_weight -p 8 -o GM12878_mustache_test.tsv

Performance

In GM12878 cells, ICE detected 15,446 loops, while Raichu identified 28,986 loops. (For this analysis, pyHICCUPS was applied; however, as shown in the manuscript, various loop-calling methods achieve a similar level of improvement when using Raichu-normalized signals.) Notably, 90.6% of loops detected by ICE (13,997 out of 15,446) were also identified by Raichu, whereas 51.7% of loops detected by Raichu (14,989 out of 28,986) were missed by ICE.

We classified the loops into three categories: ICE-specific loops, Raichu-specific loops, and common loops (detected by both ICE and Raichu). Interestingly, while ICE-specific and Raichu-specific loops showed comparable enrichment for CTCF and RAD21, Raichu-specific loops exhibited substantially greater enrichment for a broader range of transcription factors (TFs) and histone modifications closely associated with transcriptional regulation. These include RNA polymerase II (POLR2A), CREB1, RELB, H3K4me3, and H3K27ac.