dryhic

Overview

dryhic is a set of tools to manipulate HiC data.

A detailed description of the method used to remove biases (a.k.a.OneD) can be found here.

The data used for the benchmak and reproducibility comparisons can be found here

Installation

You can install the package using the handy devtools::install_github. It's highly recommended to install also the accompanying dryhicdata package, containing some useful data.

install.packages("devtools")

devtools::install_github("qenvio/dryhic")
devtools::install_github("qenvio/dryhicdata")

Alternatively, you can download, unzip and install the package manually (only UNIX)

wget https://github.com/qenvio/dryhic/archive/master.zip
unzip master.zip
mv dryhic-master dryhic
sudo R CMD INSTALL dryhic

rm master.zip

wget https://github.com/qenvio/dryhicdata/archive/master.zip
unzip master.zip
mv dryhicdata-master dryhicdata
sudo R CMD INSTALL dryhicdata

Usage

Get the data

First of all, we load the packages and some data

# dependencies

library("dplyr")
library("Matrix")
library("mgcv")

library("dryhic")
library("dryhicdata")

# load a sample matrix

data(mat)

str(mat)

By definition, a HiC contact matrix is symmetrical, so the object stores only the upper diagonal. We can symmetrize it easily

mat[1:10, 1:10]

mat <- symmetrize_matrix(mat)

mat[1:10, 1:10]

Besides the contact matrix itself, we need also some genomic information

# load some genomic information

data(bias_hg38)
data(enzymes_hg38)

str(bias_hg38)
str(enzymes_hg38)

The experiment was performed using HindIII restriction enzyme, so we gather this information

# get genomic information

info <- mutate(enzymes_hg38,
               res = HindIII) %>%
        dplyr::select(chr, pos, res) %>%
		inner_join(bias_hg38) %>%
		mutate(bin = paste0(chr, ":", pos))

summary(info)

As a sanity check, we should be sure that both the contact matrix and the genomic information refer to the very same genomic loci

common_bins <- intersect(info$bin, rownames(mat))

# watch out! this step orders chromosomes alphabetically

info <- filter(info, bin %in% common_bins) %>%
        arrange(chr, pos)

i <- match(info$bin, rownames(mat))

mat <- mat[i, i]

Now we can compute the total coverage per bin and the proportion of non-zero entries

info$tot <- Matrix::rowSums(mat)
info$nozero <- Matrix::rowMeans(mat != 0)

Filter out problematic bins

Some loci in the genome have a very poor coverage. We can filter them out based both on the HiC matrix (namely, all bins without any coverage and those presenting a very high proportion of void cells). We can further filter out bins with low mappability and with no restriction enzyme sites.

info <- filter(info,
               map > .5,
			   res > 0,
			   tot > 0,
			   nozero > .05 * median(nozero))

i <- match(info$bin, rownames(mat))

mat <- mat[i, i]

Graphical representation

In order to have a look at the data, we can select a region and create a contact map.

bw <- colorRampPalette(c("white", "black"))

bins_chr17 <- which(info$chr == "chr17")

mat_chr17 <- mat[bins_chr17, bins_chr17]

logfinite(mat_chr17) %>% image(useRaster = TRUE, main = "RAW data",
                               col.regions = bw(256), colorkey = FALSE)

Bias removal

We can apply the ICE bias correction

mat_ice <- ICE(mat, 30)

ice_chr17 <- mat_ice[bins_chr17, bins_chr17]

logfinite(ice_chr17) %>% image(useRaster = TRUE, main = "ICE",
                               col.regions = bw(256), colorkey = FALSE)

Or we can apply oned correction

info$oned <- oned(info)

mat_oned <- correct_mat_from_b(mat, info$oned)

oned_chr17 <- mat_oned[bins_chr17, bins_chr17]

logfinite(oned_chr17) %>% image(useRaster = TRUE, main = "oned",
                                col.regions = bw(256), colorkey = FALSE)

Copy number estimation

Once OneD bias removal has been applies, we can uset this result to estimate the number of copies

# get total number of bias-corrected contacts per bin

info$tot_oned <- rowSums(mat_oned)[info$bin]

# estimate CN 

info$cn <- fitcnv(info$tot_oned)[[2]]

# apply correction (the 1 / 2 factor is applied because most of the genome is diploid)

mat_onedcn <- correct_mat_from_b(mat_oned, sqrt(info$cn / 2))
info$tot_onedcn <- rowSums(mat_onedcn)[info$bin]

We can plot the total number of contacts per bin to check bias removal and CN normalization

with(info[bins_chr17,], plot(pos / 1e6, tot,
                             type = "l", las = 1, col = rgb(0, 0, 0, .5),
							 bty = "l",
							 xlab = "Genomic position / Mbp",
							 ylab = "Number of contacts per bin"))

with(info[bins_chr17,], lines(pos / 1e6, tot_oned,
                              col = rgb(1, 0, 0, .5)))

with(info[bins_chr17,], lines(pos / 1e6, tot_onedcn,
                              col = rgb(0, 0, 1, .5)))

legend("topleft", legend = c("raw", "OneD", "OneD + CN"), lty = 1,
       col = c("black", "red", "blue"), bty = "n")