extraChIPs: Differential Signal Using Fixed-Width Windows
9 July 2023
Package
extraChIPs 1.4.6
Contents
knitr::opts_chunk$set(message = FALSE, crop = NULL)1 Introduction
The GRAVI workflow, for which this
package is designed, uses sliding windows for differential signal
analysis. However, the use of fixed-width windows, as is common under
DiffBind-style (Ross-Innes et al. 2012) approaches is also possible with
extraChIPs. This vignette focusses on using conventional peak calls
and fixed-width approaches to replicate and extend these approaches.
The majority of examples below use heavily reduced datasets to provide general guidance on using the functions. Some results may appear trivial as a result, but will hopefully prove far more useful in a true experimental context. All data, along with this vignette are available here. Please place all contents of the data directory in a directory named data in your own working directory.
2 Setup
2.1 Installation
In order to use the package extraChIPs and follow this vignette, we
recommend using the package BiocManager hosted on CRAN. Once this is
installed, the additional packages required for this vignette
(tidyverse, Rsamtools, csaw, BiocParallel and rtracklayer) can
also be installed.
if (!"BiocManager" %in% rownames(installed.packages()))
install.packages("BiocManager")
pkg <- c(
"tidyverse", "Rsamtools", "csaw", "BiocParallel", "rtracklayer", "edgeR",
"patchwork", "extraChIPs", "plyranges", "scales", "here", "quantro"
)
BiocManager::install(pkg, update = FALSE)Once these packages are installed, we can load them easily
library(tidyverse)
library(Rsamtools)
library(csaw)
library(BiocParallel)
library(rtracklayer)
library(edgeR)
library(patchwork)
library(extraChIPs)
library(plyranges)
library(scales)
library(glue)
library(ggrepel)
library(here)
library(quantro)
theme_set(theme_bw())2.2 Data
All data for this vignette is expected to be in a sub-directory of the working directory named “data”, and all paths will be predicated on this. Please ensure you have all data in this location, obtained from here.
The data itself is ChIP-Seq data targeting the Estrogen Receptor (ER),
and is taken from the cell-line ZR-75-1 cell-line using data from the
BioProject , Pre-processing was performed using the
prepareChIPs workflow,
written in snakemake (Mölder et al. 2021) and all code is available at
https://github.com/smped/PRJNA509779. ER binding was assessed under
Vehicle (E2) and DHT-stimulated (E2DHT) conditions. Using GRCh37 as the
reference genome, a subset of regions found on chromosome 10 are
included in this dataset for simplicity.
First we’ll define our sample data then define our two treatment groups. Defining a consistent colour palette for all plots is also a good habit to develop.
treat_levels <- c("E2", "E2DHT")
treat_colours <- setNames(c("steelblue", "red3"), treat_levels)
samples <- tibble(
accession = paste0("SRR831518", seq(0, 5)),
target = "ER",
treatment = factor(rep(treat_levels, each = 3), levels = treat_levels)
)
samples## # A tibble: 6 × 3
## accession target treatment
## <chr> <chr> <fct>
## 1 SRR8315180 ER E2
## 2 SRR8315181 ER E2
## 3 SRR8315182 ER E2
## 4 SRR8315183 ER E2DHT
## 5 SRR8315184 ER E2DHT
## 6 SRR8315185 ER E2DHTaccessions <- samples %>%
split(f = .$treatment) %>%
lapply(pull, "accession")We’ll eventually be loading counts for differential signal analysis from
a set of BamFiles, so first we’ll create a BamFileList with all of
these files.
bfl <- here("data", "ER", glue("{samples$accession}.bam")) %>%
BamFileList() %>%
setNames(str_remove_all(names(.), ".bam"))
file.exists(path(bfl))## [1] TRUE TRUE TRUE TRUE TRUE TRUEThis also enables creation of a Seqinfo object based on
the actual reference genome to which the reads were aligned during data
preparation. Seqinfo objects are the foundation of working with
GRanges, so defining an object at the start of a workflow is good
practice.
As is common practice for ChIP-Seq analyses, we’ll restrict our focus to the
standard chromosomes.
sq <- seqinfo(bfl)
sq <- keepStandardChromosomes(sq)
isCircular(sq) <- rep(FALSE, length(seqlevels(sq)))
genome(sq) <- "GRCh37"Another key preparatory step for working with peaks is to define a set of regions as either blacklisted or grey-listed regions. The former are known problematic regions based on each genome, with data freely available from https://github.com/Boyle-Lab/Blacklist/tree/master/lists, whilst grey-listed regions are defined from potentially problematic regions as detected within the input sample. For our samples code for this is included in the previously provided repository (https://github.com/smped/PRJNA509779).
greylist <- import.bed(here("data/chr10_greylist.bed"), seqinfo = sq)
blacklist <- import.bed( here("data/chr10_blacklist.bed"), seqinfo = sq)
omit_ranges <- c(greylist, blacklist)3 Working With Peaks
The provided dataset includes six files produced by macs2 callpeak
(Zhang et al. 2008) in the narrowPeak format, and these are able to be
easily parsed using extraChIPs. We’ll immediately pass our black &
grey-listed regions to our parsing function so we can exclude these
regions right from the start.
By passing the above seqinfo object to this function, we’re also telling
importPeaks() to ignore any peaks not on the included chromosomes.
peaks <- here("data", "ER", glue("{samples$accession}_peaks.narrowPeak")) %>%
importPeaks(seqinfo = sq, blacklist = omit_ranges)This will import the peaks from all files as a single GRangesList
object, adding the file-name to each element by default. We can easily
modify these names if we so wish.
names(peaks) <- str_remove_all(names(peaks), "_peaks.narrowPeak")Once loaded, we can easily check how similar our replicates are using
plotOverlaps(). When three or more sets of peaks are contained in the
GRangesList, an UpSet plot will be drawn by default.
plotOverlaps(
peaks, min_size = 10, .sort_sets = FALSE,
set_col = treat_colours[as.character(samples$treatment)]
)Optionally, specifying a column and a suitable function will produce an additional panel summarising that value. In the following, we’ll show the maximum score obtained, highlighting that for peaks identified in only one or two replicates, the overall signal intensity is generally lower, even in the sample with the strongest signal.
plotOverlaps(
peaks, min_size = 10, .sort_sets = FALSE, var = "score", f = "max",
set_col = treat_colours[as.character(samples$treatment)]
)
Figure 1: UpSet plot showing overlapping peaks across all replicates, with the maximum score across all replicates shown in the upper panel.
A common task at this point may be to define consensus peaks within each
treatment group, by retaining only the peaks found in 2 of the 3
replicates (p = 2/3). The default approach is to take the union of all
ranges, with the returned object containing logical values for each
sample, as well as the number of samples where an overlapping peak was
found.
If we wish to retain any of the original columns, such as the
macs2 callpeak score, we can simply pass the column names to
makeConsensus()
consensus_e2 <- makeConsensus(peaks[accessions$E2], p = 2/3, var = "score")
consensus_e2dht <- makeConsensus(peaks[accessions$E2DHT], p = 2/3, var = "score")Alternatively, we could find the centre of the peaks as part of this process, by averaging across the estimated peak centres for each sample. This is a very common step for ChIP-Seq data where the target is a transcription factor, and also forms a key step in the DiffBind workflow.
In the following code chunk, we first find the centre for each sample
using the information provided by macs2, before retaining this column
when calling makeConsensus(). This will return each of the individual
centre-position estimates as a list for each merged range, and using
vapply() we then take the mean position as our estimate for the
combined peak centre.
consensus_e2 <- peaks[accessions$E2] %>%
endoapply(mutate, centre = start + peak) %>%
makeConsensus(p = 2/3, var = "centre") %>%
mutate(centre = vapply(centre, mean, numeric(1)))
consensus_e2## GRanges object with 164 ranges and 5 metadata columns:
## seqnames ranges strand | centre SRR8315180 SRR8315181
## <Rle> <IRanges> <Rle> | <numeric> <logical> <logical>
## [1] chr10 43048195-43048529 * | 43048362 TRUE TRUE
## [2] chr10 43521739-43522260 * | 43522020 TRUE TRUE
## [3] chr10 43540042-43540390 * | 43540272 TRUE FALSE
## [4] chr10 43606238-43606573 * | 43606416 TRUE TRUE
## [5] chr10 43851214-43851989 * | 43851719 FALSE TRUE
## ... ... ... ... . ... ... ...
## [160] chr10 99096784-99097428 * | 99097254 TRUE TRUE
## [161] chr10 99168353-99168649 * | 99168502 TRUE TRUE
## [162] chr10 99207868-99208156 * | 99207998 FALSE TRUE
## [163] chr10 99331363-99331730 * | 99331595 TRUE TRUE
## [164] chr10 99621632-99621961 * | 99621818 FALSE TRUE
## SRR8315182 n
## <logical> <numeric>
## [1] TRUE 3
## [2] TRUE 3
## [3] TRUE 2
## [4] TRUE 3
## [5] TRUE 2
## ... ... ...
## [160] TRUE 3
## [161] TRUE 3
## [162] TRUE 2
## [163] TRUE 3
## [164] TRUE 2
## -------
## seqinfo: 25 sequences from GRCh37 genomeconsensus_e2dht <- peaks[accessions$E2DHT] %>%
endoapply(mutate, centre = start + peak) %>%
makeConsensus(p = 2/3, var = "centre") %>%
mutate(centre = vapply(centre, mean, numeric(1)))We can also inspect these using plotOverlaps() provided we use a
GRangesList for the input. Now that we only have two elements (one for
each treatment) a VennDiagram will be generated instead of an UpSet
plot.
GRangesList(E2 = granges(consensus_e2), E2DHT = granges(consensus_e2dht)) %>%
plotOverlaps(set_col = treat_colours[treat_levels])
Figure 2: Overlap between consensus peaks identified in a treatment-specific manner
We can now go one step further and define the set of peaks found in either treatment. Given we’re being inclusive here, we can leave p = 0 so any peak found in either treatment is included.
union_peaks <- GRangesList(
E2 = select(consensus_e2, centre),
E2DHT = select(consensus_e2dht, centre)
) %>%
makeConsensus(var = c("centre")) %>%
mutate(
centre = vapply(centre, mean, numeric(1)) %>% round(0)
) Now we have a set of peaks, found in at least 2/3 of samples from either condition, with estimates of each peak’s centre. The next step would be to set all peaks as the same width based on the centre position, with a common width being 500bp.
In the following we’ll perform multiple operations in a single call mutate, so let’s make sure we know what’s happening.
glue("{seqnames}:{centre}:{strand}")usesgluesyntax to parse the seqnames, centre position and strand information as a character-like vector with a width of only 1, and using the estimated centre as the Range.- We then coerce this to a
GRangesobject, before resizing to the desired width. - We also add the original (un-centred) range as an additional column,
retaining the
GRangesstructure, but discarding anything in themcols()element, then - Using
colToRanges(), we take the centred ranges and place them as the core set of GRanges for this object.
This gives a GRanges object with all original information, but with centred peaks of a fixed width.
w <- 500
centred_peaks <- union_peaks %>%
mutate(
centre = glue("{seqnames}:{centre}:{strand}") %>%
GRanges(seqinfo = sq) %>%
resize(width = w),
union_peak = granges(.)
) %>%
colToRanges("centre")4 Counting Reads
Now we have our centred, fixed-width peaks, we can count reads using
csaw::regionCounts() (Lun and Smyth 2016). We know our fragment length
is about 200bp, so we can pass this to the function for a slightly more
sophisticated approach to counting.
se <- regionCounts(bfl, centred_peaks, ext = 200)se## class: RangedSummarizedExperiment
## dim: 188 6
## metadata(2): final.ext param
## assays(1): counts
## rownames: NULL
## rowData names(4): E2 E2DHT n union_peak
## colnames(6): SRR8315180 SRR8315181 ... SRR8315184 SRR8315185
## colData names(4): bam.files totals ext rlenThe colData() element of the returned object as the columns
bam.files, totals, ext and rlen, which are all informative and
can be supplemented with our samples data frame. In the following,
we’ll 1) coerce to a tibble, 2) left_join() the samples object, 3)
add the accession as the sample column, 4) set the accession back as the
rownames, then 5) coerce back to the required DataFrame() structure.
colData(se) <- colData(se) %>%
as_tibble(rownames = "accession") %>%
left_join(samples) %>%
mutate(sample = accession) %>%
as.data.frame() %>%
column_to_rownames("accession") %>%
DataFrame()
colData(se)## DataFrame with 6 rows and 7 columns
## bam.files totals ext rlen target
## <character> <integer> <integer> <integer> <character>
## SRR8315180 /home/steviep/github.. 317845 200 75 ER
## SRR8315181 /home/steviep/github.. 337623 200 75 ER
## SRR8315182 /home/steviep/github.. 341998 200 75 ER
## SRR8315183 /home/steviep/github.. 315872 200 75 ER
## SRR8315184 /home/steviep/github.. 352908 200 75 ER
## SRR8315185 /home/steviep/github.. 347709 200 75 ER
## treatment sample
## <factor> <character>
## SRR8315180 E2 SRR8315180
## SRR8315181 E2 SRR8315181
## SRR8315182 E2 SRR8315182
## SRR8315183 E2DHT SRR8315183
## SRR8315184 E2DHT SRR8315184
## SRR8315185 E2DHT SRR8315185For QC and visualisation, we can add an additional logCPM assay to our
object as well.
assay(se, "logCPM") <- cpm(assay(se, "counts"), lib.size = se$totals, log = TRUE)First we might like to check our distribution of counts
plotAssayDensities(se, assay = "counts", colour = "treat", trans = "log1p") +
scale_colour_manual(values = treat_colours)
Figure 3: Count densities for all samples, using the log+1 transformation
A PCA plot can also provide insight as to where the variability in the data lies.
plotAssayPCA(se, assay = "logCPM", colour = "treat", label = "sample") +
scale_colour_manual(values = treat_colours)
Figure 4: PCA plot using logCPM values and showing that replicate variability is larger than variability between treatment groups.
5 Differential Signal Analysis
5.1 Statistical Testing
In order to perform Differential Signal Analysis, we simply need to
define a model matrix, as for conventional analysis using edgeR or
limma. We can then pass this, along with our fixed-width counts to
fitAssayDiff(). By default normalisation will be library-size
normalisation, as is a common default strategy for ChIP-Seq data. In
contrast to sliding window approaches, these results represent our final
results and there is no need for merging windows.
X <- model.matrix(~treatment, data = colData(se))
ls_res <- fitAssayDiff(se, design = X, asRanges = TRUE)
sum(ls_res$FDR < 0.05)## [1] 3TMM normalisation (Robinson and Oshlack 2010) is another common
strategy, which relies on the data from all treatment groups being drawn
from the same distributions. We can formally test this using the package
quantro (Hicks and Irizarry 2015) , which produces p-values for 1)
H0: Group medians are drawn from the same distribution, and
2) H0: Group-specific distributions are the same.
set.seed(100)
qtest <- assay(se, "counts") %>%
quantro(groupFactor = se$treatment, B = 1e3)
qtest## quantro: Test for global differences in distributions
## nGroups: 2
## nTotSamples: 6
## nSamplesinGroups: 3 3
## anovaPval: 0.90754
## quantroStat: 0.21859
## quantroPvalPerm: 0.572Here, both p-values are >0.05, so in conjunction with out visual
inspection earlier, we can confidently apply TMM normalisation. To apply
this, we simply specify the argument norm = "TMM" when we call
fitAssayDiff(). In the analysis below, we’ve also specified a
fold-change threshold (fc = 1.2), below which, changes in signal are
considered to not be of interest (McCarthy and Smyth 2009). This
threshold is incorporated into the testing so there is no requirement
for post-hoc filtering based on a threshold.
tmm_res <- fitAssayDiff(se, design = X, norm = "TMM", asRanges = TRUE, fc = 1.2)
sum(tmm_res$FDR < 0.05)## [1] 7An MA-plot is a common way of inspecting results and in the following we use the original ‘union_peak’ in our labelling of points. This serves as a reminder that the fixed-width windows are in fact a proxy for the entire region for which we have confidently detected ChIP signal, and that these windows are truly the regions of interest.
tmm_res %>%
as_tibble() %>%
mutate(`FDR < 0.05` = FDR < 0.05) %>%
ggplot(aes(logCPM, logFC)) +
geom_point(aes(colour = `FDR < 0.05`)) +
geom_smooth(se = FALSE) +
geom_label_repel(
aes(label = union_peak), colour = "red",
data = . %>% dplyr::filter(FDR < 0.05)
) +
scale_colour_manual(values = c("black", "red"))
Figure 5: MA-plot after fitting using TMM normalisation and applying a fold-change threshold during testing
Points are labelled using the original windows obtained when merging replicates and treatment groups.
5.2 Mapping to Genes
Whilst knowledge of which regions are showing differential signal, the fundamental question we are usually asking is about the downstream regulatory consequences, such as the target gene. Before we can map peaks to genes, we’ll need to define our genes. In the following, we’ll use the provided Gencode gene mappings at the gene, transcript and exon level.
gencode <- here("data/gencode.v43lift37.chr10.annotation.gtf.gz") %>%
import.gff() %>%
filter_by_overlaps(GRanges("chr10:42354900-100000000")) %>%
split(.$type)
seqlevels(gencode) <- seqlevels(sq)
seqinfo(gencode) <- sqMapping to genes using mapByFeature() uses additional annotations,
such as whether the peak overlaps a promoter, enhancer or long-range
interaction. Here we’ll just use promoters, so let’s create a set of
promoters from our transcript-level information, ensuring we incorporate
all possible promoters within a gene, and merging any overlapping ranges
using reduceMC()
promoters <- gencode$transcript %>%
select(gene_id, ends_with("name")) %>%
promoters(upstream = 2500, downstream = 500) %>%
reduceMC(simplify = FALSE)
promoters## GRanges object with 1678 ranges and 3 metadata columns:
## seqnames ranges strand |
## <Rle> <IRanges> <Rle> |
## [1] chr10 42678287-42681286 + |
## [2] chr10 42702938-42705937 + |
## [3] chr10 42735669-42738668 + |
## [4] chr10 42743933-42746932 + |
## [5] chr10 42968428-42973155 + |
## ... ... ... ... .
## [1674] chr10 99635155-99638154 - |
## [1675] chr10 99643500-99646805 - |
## [1676] chr10 99695536-99698535 - |
## [1677] chr10 99770595-99773594 - |
## [1678] chr10 99789879-99793085 - |
## gene_id
## <CharacterList>
## [1] ENSG00000237592.2_5
## [2] ENSG00000271650.1_7
## [3] ENSG00000290458.1_2
## [4] ENSG00000274167.5_8
## [5] ENSG00000185904.12_9,ENSG00000185904.12_9,ENSG00000185904.12_9,...
## ... ...
## [1674] ENSG00000265398.1
## [1675] ENSG00000095713.14_13,ENSG00000095713.14_13
## [1676] ENSG00000095713.14_13
## [1677] ENSG00000095713.14_13
## [1678] ENSG00000095713.14_13,ENSG00000095713.14_13
## gene_name
## <CharacterList>
## [1] IGKV1OR10-1
## [2] ENSG00000271650
## [3] ENSG00000290458
## [4] ENSG00000274167
## [5] LINC00839,LINC00839,LINC00839,...
## ... ...
## [1674] AL139239.1
## [1675] CRTAC1,CRTAC1
## [1676] CRTAC1
## [1677] CRTAC1
## [1678] CRTAC1,CRTAC1
## transcript_name
## <CharacterList>
## [1] IGKV1OR10-1-201
## [2] ENST00000605702
## [3] ENST00000622823
## [4] ENST00000622650
## [5] LINC00839-204,LINC00839-203,LINC00839-202,...
## ... ...
## [1674] AL139239.1-201
## [1675] CRTAC1-205,CRTAC1-206
## [1676] CRTAC1-204
## [1677] CRTAC1-201
## [1678] CRTAC1-203,CRTAC1-202
## -------
## seqinfo: 25 sequences from GRCh37 genomeNow we’ll pass these to mapByFeature(), but first, we’ll place the
original ‘union_peak’ back as the core of the GRanges object. This will
retain all the results from testing, but ensures the correct region is
mapped to genes.
tmm_mapped_res <- tmm_res %>%
colToRanges("union_peak") %>%
mapByFeature(genes = gencode$gene, prom = promoters) %>%
mutate(
status = case_when(
FDR >= .05 ~ "Unchanged",
logFC > 0 ~ "Increased",
logFC < 0 ~ "Decreased"
)
)
arrange(tmm_mapped_res, PValue)## GRanges object with 188 ranges and 10 metadata columns:
## seqnames ranges strand | E2 E2DHT n
## <Rle> <IRanges> <Rle> | <logical> <logical> <numeric>
## [1] chr10 81101906-81102928 * | TRUE TRUE 2
## [2] chr10 79629641-79630271 * | TRUE TRUE 2
## [3] chr10 89407752-89408138 * | FALSE TRUE 1
## [4] chr10 52233596-52233998 * | TRUE TRUE 2
## [5] chr10 91651138-91651433 * | FALSE TRUE 1
## ... ... ... ... . ... ... ...
## [184] chr10 89395292-89395626 * | TRUE TRUE 2
## [185] chr10 82723984-82724328 * | TRUE TRUE 2
## [186] chr10 93120411-93121224 * | TRUE TRUE 2
## [187] chr10 95755308-95755721 * | TRUE TRUE 2
## [188] chr10 79190987-79191351 * | FALSE TRUE 1
## logFC logCPM PValue FDR
## <numeric> <numeric> <numeric> <numeric>
## [1] 1.880493 7.91492 2.50406e-25 4.70764e-23
## [2] 0.940221 8.07204 8.07816e-08 7.59347e-06
## [3] 1.578056 6.16574 4.18078e-07 2.61995e-05
## [4] 1.057814 6.67466 5.73897e-05 2.69732e-03
## [5] 1.546530 5.15431 1.75562e-04 6.60112e-03
## ... ... ... ... ...
## [184] -0.01346141 6.36140 0.961749 0.982656
## [185] -0.00855862 6.91968 0.972422 0.988154
## [186] -0.01173559 9.32397 0.979561 0.988154
## [187] 0.00721240 5.88042 0.982898 0.988154
## [188] -0.00157801 6.41182 0.994914 0.994914
## gene_id
## <CharacterList>
## [1] ENSG00000108179.14_6
## [2] ENSG00000151208.17_11
## [3] ENSG00000225913.2_9,ENSG00000196566.2_10
## [4] ENSG00000198964.14_10,ENSG00000225303.2_10
## [5] ENSG00000280560.3_9
## ... ...
## [184] ENSG00000225913.2_9,ENSG00000196566.2_10
## [185] ENSG00000227209.1_5
## [186] ENSG00000289228.2_2
## [187] ENSG00000138193.17_12
## [188] ENSG00000156113.25_17
## gene_name status
## <CharacterList> <character>
## [1] PPIF Increased
## [2] DLG5 Increased
## [3] ENSG00000225913,ENSG00000196566 Increased
## [4] SGMS1,ENSG00000225303 Increased
## [5] LINC01374 Increased
## ... ... ...
## [184] ENSG00000225913,ENSG00000196566 Unchanged
## [185] WARS2P1 Unchanged
## [186] ENSG00000289228 Unchanged
## [187] PLCE1 Unchanged
## [188] KCNMA1 Unchanged
## -------
## seqinfo: 1 sequence from an unspecified genome; no seqlengths6 Inspection of Results
6.1 Profile Heatmaps
When analysing a transcription factor, checking the binding profile across our treatment groups can be informative, and is often performed using ‘Profile Heatmaps’ where coverage is smoothed within bins surrounding our peak centre.
The function getProfileData() takes a set of ranges and a
BigWigFileList, and performs the smoothing, which is then passed to the
function plotProfileHeatmap().
The following shows the three steps of 1) defining the ranges, 2) obtaining the smoothed binding profiles, and 3) drawing the heatmap. Note that we can facet the heatmaps by selecting the ‘status’ column to separate any Increased or Decreased regions. By default, this will also draw the smoothed lines in the top panel using different colours.
These plots can be used to show coverage-like values (SPMR or CPM) or we
can use fold-enrichment over the input sample(s), as is also produced by
macs2 bdgcmp. This data isn’t generally visualised using
log-transformation so we’ll set log = FALSE in our call to
getProfileData()
fe_bw <- here("data", "ER", glue("{treat_levels}_FE_chr10.bw")) %>%
BigWigFileList() %>%
setNames(treat_levels)
sig_ranges <- filter(tmm_mapped_res, FDR < 0.05)
pd_fe <- getProfileData(fe_bw, sig_ranges, log = FALSE)
pd_fe %>%
plotProfileHeatmap("profile_data", facetY = "status") +
scale_fill_gradient(low = "white", high = "red") +
labs(fill = "Fold\nEnrichment")6.2 Session Info
sessionInfo()## R version 4.3.1 (2023-06-16)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 22.04.2 LTS
##
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.17-bioc/R/lib/libRblas.so
## LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.10.0
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_GB LC_COLLATE=C
## [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## time zone: America/New_York
## tzcode source: system (glibc)
##
## attached base packages:
## [1] stats4 stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] quantro_1.34.0 here_1.0.1
## [3] ggrepel_0.9.3 glue_1.6.2
## [5] scales_1.2.1 plyranges_1.20.0
## [7] extraChIPs_1.4.6 patchwork_1.1.2
## [9] edgeR_3.42.4 limma_3.56.2
## [11] rtracklayer_1.60.0 BiocParallel_1.34.2
## [13] csaw_1.34.0 SummarizedExperiment_1.30.2
## [15] Biobase_2.60.0 MatrixGenerics_1.12.2
## [17] matrixStats_1.0.0 Rsamtools_2.16.0
## [19] Biostrings_2.68.1 XVector_0.40.0
## [21] GenomicRanges_1.52.0 GenomeInfoDb_1.36.1
## [23] IRanges_2.34.1 S4Vectors_0.38.1
## [25] BiocGenerics_0.46.0 lubridate_1.9.2
## [27] forcats_1.0.0 stringr_1.5.0
## [29] dplyr_1.1.2 purrr_1.0.1
## [31] readr_2.1.4 tidyr_1.3.0
## [33] tibble_3.2.1 ggplot2_3.4.2
## [35] tidyverse_2.0.0 BiocStyle_2.28.0
##
## loaded via a namespace (and not attached):
## [1] ProtGenerics_1.32.0 bitops_1.0-7
## [3] httr_1.4.6 RColorBrewer_1.1-3
## [5] doParallel_1.0.17 InteractionSet_1.28.1
## [7] tools_4.3.1 doRNG_1.8.6
## [9] backports_1.4.1 utf8_1.2.3
## [11] R6_2.5.1 HDF5Array_1.28.1
## [13] mgcv_1.8-42 lazyeval_0.2.2
## [15] Gviz_1.44.0 rhdf5filters_1.12.1
## [17] GetoptLong_1.0.5 withr_2.5.0
## [19] prettyunits_1.1.1 gridExtra_2.3
## [21] base64_2.0.1 VennDiagram_1.7.3
## [23] preprocessCore_1.62.1 cli_3.6.1
## [25] formatR_1.14 labeling_0.4.2
## [27] sass_0.4.6 genefilter_1.82.1
## [29] askpass_1.1 foreign_0.8-84
## [31] siggenes_1.74.0 illuminaio_0.42.0
## [33] dichromat_2.0-0.1 scrime_1.3.5
## [35] BSgenome_1.68.0 rstudioapi_0.15.0
## [37] RSQLite_2.3.1 generics_0.1.3
## [39] shape_1.4.6 BiocIO_1.10.0
## [41] Matrix_1.6-0 interp_1.1-4
## [43] futile.logger_1.4.3 fansi_1.0.4
## [45] lifecycle_1.0.3 yaml_2.3.7
## [47] rhdf5_2.44.0 BiocFileCache_2.8.0
## [49] grid_4.3.1 blob_1.2.4
## [51] crayon_1.5.2 lattice_0.21-8
## [53] ComplexUpset_1.3.3 GenomicFeatures_1.52.1
## [55] annotate_1.78.0 KEGGREST_1.40.0
## [57] pillar_1.9.0 knitr_1.43
## [59] ComplexHeatmap_2.16.0 beanplot_1.3.1
## [61] metapod_1.8.0 rjson_0.2.21
## [63] codetools_0.2-19 data.table_1.14.8
## [65] vctrs_0.6.3 png_0.1-8
## [67] gtable_0.3.3 cachem_1.0.8
## [69] xfun_0.39 S4Arrays_1.0.4
## [71] ggside_0.2.2 survival_3.5-5
## [73] iterators_1.0.14 GenomicInteractions_1.34.0
## [75] nlme_3.1-162 bit64_4.0.5
## [77] progress_1.2.2 filelock_1.0.2
## [79] rprojroot_2.0.3 bslib_0.5.0
## [81] nor1mix_1.3-0 rpart_4.1.19
## [83] colorspace_2.1-0 DBI_1.1.3
## [85] Hmisc_5.1-0 nnet_7.3-19
## [87] tidyselect_1.2.0 bit_4.0.5
## [89] compiler_4.3.1 curl_5.0.1
## [91] htmlTable_2.4.1 xml2_1.3.5
## [93] DelayedArray_0.26.6 bookdown_0.34
## [95] checkmate_2.2.0 quadprog_1.5-8
## [97] rappdirs_0.3.3 digest_0.6.33
## [99] rmarkdown_2.23 GEOquery_2.68.0
## [101] htmltools_0.5.5 pkgconfig_2.0.3
## [103] jpeg_0.1-10 base64enc_0.1-3
## [105] sparseMatrixStats_1.12.2 highr_0.10
## [107] dbplyr_2.3.3 fastmap_1.1.1
## [109] ensembldb_2.24.0 rlang_1.1.1
## [111] GlobalOptions_0.1.2 htmlwidgets_1.6.2
## [113] DelayedMatrixStats_1.22.2 EnrichedHeatmap_1.30.0
## [115] farver_2.1.1 jquerylib_0.1.4
## [117] jsonlite_1.8.7 mclust_6.0.0
## [119] VariantAnnotation_1.46.0 RCurl_1.98-1.12
## [121] magrittr_2.0.3 Formula_1.2-5
## [123] GenomeInfoDbData_1.2.10 Rhdf5lib_1.22.0
## [125] munsell_0.5.0 Rcpp_1.0.11
## [127] stringi_1.7.12 zlibbioc_1.46.0
## [129] MASS_7.3-60 plyr_1.8.8
## [131] bumphunter_1.42.0 minfi_1.46.0
## [133] parallel_4.3.1 deldir_1.0-9
## [135] splines_4.3.1 multtest_2.56.0
## [137] hms_1.1.3 circlize_0.4.15
## [139] locfit_1.5-9.8 igraph_1.5.0
## [141] rngtools_1.5.2 biomaRt_2.56.1
## [143] futile.options_1.0.1 XML_3.99-0.14
## [145] evaluate_0.21 latticeExtra_0.6-30
## [147] biovizBase_1.48.0 lambda.r_1.2.4
## [149] BiocManager_1.30.21 tzdb_0.4.0
## [151] foreach_1.5.2 tweenr_2.0.2
## [153] openssl_2.0.6 polyclip_1.10-4
## [155] reshape_0.8.9 clue_0.3-64
## [157] ggforce_0.4.1 broom_1.0.5
## [159] xtable_1.8-4 restfulr_0.0.15
## [161] AnnotationFilter_1.24.0 memoise_2.0.1
## [163] AnnotationDbi_1.62.2 GenomicAlignments_1.36.0
## [165] cluster_2.1.4 timechange_0.2.06.3 References
Hahne, Florian, and Robert Ivanek. 2016. “Statistical Genomics: Methods and Protocols.” In, edited by Ewy Mathé and Sean Davis, 335–51. New York, NY: Springer New York. https://doi.org/10.1007/978-1-4939-3578-9_16.
Hicks, Stephanie C, and Rafael A Irizarry. 2015. “Quantro: A Data-Driven Approach to Guide the Choice of an Appropriate Normalization Method.” Genome Biol. 16 (1): 117.
Lun, Aaron T L, and Gordon K Smyth. 2016. “Csaw: A Bioconductor Package for Differential Binding Analysis of ChIP-Seq Data Using Sliding Windows.” Nucleic Acids Res. 44 (5): e45.
McCarthy, Davis J, and Gordon K Smyth. 2009. “Testing Significance Relative to a Fold-Change Threshold Is a TREAT.” Bioinformatics 25 (6): 765–71.
Mölder, Felix, Kim Philipp Jablonski, Brice Letcher, Michael B Hall, Christopher H Tomkins-Tinch, Vanessa Sochat, Jan Forster, et al. 2021. “Sustainable Data Analysis with Snakemake.” F1000Res. 10 (January): 33.
Robinson, Mark D, and Alicia Oshlack. 2010. “A Scaling Normalization Method for Differential Expression Analysis of RNA-seq Data.” Genome Biol. 11 (3): R25.
Ross-Innes, Caryn S., Rory Stark, Andrew E. Teschendorff, Kelly A. Holmes, H. Raza Ali, Mark J. Dunning, Gordon D. Brown, et al. 2012. “Differential Oestrogen Receptor Binding Is Associated with Clinical Outcome in Breast Cancer.” Nature 481: –4. http://www.nature.com/nature/journal/v481/n7381/full/nature10730.html.
Zhang, Yong, Tao Liu, Clifford A Meyer, Jérôme Eeckhoute, David S Johnson, Bradley E Bernstein, Chad Nusbaum, et al. 2008. “Model-Based Analysis of ChIP-Seq (MACS).” Genome Biol. 9 (9): R137.