recountmethylation 1.4.5
The recountmethylation package provides access to databases of DNA
methylation (DNAm) data from over 62,000 cumulative sample records with
IDATs in the Gene Expression Omnibus (GEO, available by November, 2020).
Samples were run using either of 2 Illumina BeadArray platform types,
either the older HM450K platform or the newer EPIC/HM850K platform. The
database compilation files include mined, mapped, and model-based sample
metadata, and DNAm data in the form of either raw/unnormalized red and
green signals, raw/unnormalized methylated and unmethylated signals, or
normalized DNAm fractions (a.k.a. “Beta-values”) [1].
Normalization was performed using the out-of-band signal correction (a.k.a.
“noob”) method, a type of within-sample normalization [2].
This User’s Guide shows how to use the recountmethylation package to obtain,
load, and query the DNAm databases with 2 small example files. Background about
DNAm arrays, DNAm measurement, SummarizedExperiment objects, database file
types, and samples metadata is also provided. Further analysis examples are
contained in the data_analyses vignette.
Databases accessed with recountmethylation contain data from GEO
(ncbi.nlm.nih.gov/geo/), a live public database where alterations to
online records can cause discrepancies with stored data over time.
We cannot guarantee the accuracy of stored data, and advise users
cross-check their findings with latest available records.
Database compilation file download and access is managed by the get_db
functions, where the DNAm array platform type using the platform argument
(see ?get_db for details). Both HM450K and EPIC/HM850K platforms are
currently supported (see below for platform details). Note you will need
between 50-180 Gb of disk space to store a single database file. Files pair
sample metadata and assay data in various formats, including HDF5-SummarizedExperiment database directories, and HDF5 database files
with the .h5 extension.
The databases are located at https://methylation.recount.bio/, and file details are viewable as follows:
sm <- as.data.frame(smfilt(get_servermatrix()))
if(is(sm, "data.frame")){knitr::kable(sm, align = "c")}| filename | date | time | size (bytes) |
|---|---|---|---|
| remethdb_h5-rg_epic_0-0-2_1589820348.h5 | 08-Jan-2021 | 09:46 | 66751358297 |
| remethdb_h5se-gm_epic_0-0-2_1589820348 | 06-Jan-2021 | 09:56 | assays.h5 = 56956363488;se.rds = 8475111 |
| remethdb_h5se-gr_epic_0-0-2_1607018051 | 06-Jan-2021 | 10:09 | assays.h5 = 82090895411;se.rds = 8475201 |
| remethdb_h5se-rg_epic_0-0-2_1589820348 | 06-Jan-2021 | 10:19 | assays.h5 = 68707689800;se.rds = 3059883 |
| remethdb_h5-rg_hm450k_0-0-2_1607018051.h5 | 07-Jan-2021 | 10:01 | 193342823766 |
| remethdb_h5se-gm_hm450k_0-0-2_1607018051 | 07-Jan-2021 | 10:23 | assays.h5 = 130935841655;se.rds = 5372091 |
| remethdb_h5se-gr_hm450k_0-0-2_1607018051 | 07-Jan-2021 | 10:50 | assays.h5 = 184355830172;se.rds = 5372008 |
| remethdb_h5se-rg_hm450k_0-0-2_1607018051 | 06-Jan-2021 | 10:44 | assays.h5 = 164788908310;se.rds = 3179962 |
| remethdb-h5se_gr-test_0-0-1_1590090412 | 29-May-2020 | 07:28 | assays.h5 = 132596;se.rds = 68522 |
| remethdb-h5_rg-test_0-0-1_1590090412.h5 | 31-May-2020 | 07:26 | 252757 |
The DNAm array database files are indexed on ExperimentHub, and are
viewable as follows. Note, the cache needs to be set with R_user_dir() per instructions here.
cache.path <- tools::R_user_dir("recountmethylation")
setExperimentHubOption("CACHE", cache.path)
hub <- ExperimentHub::ExperimentHub() # connect to the hubs
rmdat <- AnnotationHub::query(hub, "recountmethylation") # query the hubsIn addition to using the getdb functions, the HDF5 (“.h5”" extension)
files may be downloaded from the hubs.
fpath <- rmdat[["EH3778"]] # download with default caching
rhdf5::h5ls(fpath) # load the h5 fileNote that whether downloads use the hubs or getdb functions, caching
is implemented to check for previously downloaded database files.
This section includes essential background about DNAm array platforms, assays and file types, and sample metadata.
Databases include human samples run on the Illumina Infinium HM450K BeadArray platform. HM450K is a popular 2-channel platform that probes over 480,000 CpG loci genome-wide, with enriched coverage at CG islands, genes, and enhancers [3]. The more recently released EPIC/HM850K platform contains an expanded probe set targeting over 850,000 CpGs, including more than 90% of the HM450K probes, with greater coverage of potential intergenic regulatory regions [4].
Array processing generates 2 intensity files (IDATs) per sample, one each for
the red and green color channels. These raw files also contain control signals
useful for quality evaluations [5]. The BeadArray probes use
either of 2 bead technologies, known as Type I and Type II, where the majority
(72%) of probes use the latter. For Type II probes, a single bead assay informs
a single probe, while Type I probes use 2 beads each. Practically, this means
the bead-specific matrices found in RGChannelSet objects are larger than the
probe-specific matrices found in derived object types (e.g. for HM450K samples,
622,399 assays for red/green signal matrices versus 485,512 assays for
methylated/unmethylated signal, DNAm fractions matrices, see below).
SummarizedExperiment object classesDNAm array sample IDATs can be read into an R session as an object of class
RGChannelSet, a type of SummarizedExperiment. These objects support
analyses of high-throughput genomics datasets, and they include slots for
assay matrices, sample metadata, and experiment metadata. During a typical
workflow, normalization and preprocessing convert RGChannelSet objects into
new types like MethylSet and RatioSet. While not all IDAT information is
accessible from every object type (e.g. only RGChannelSets can contain
control assays), derived objects like MethylSets and RatioSets may be
smaller and/or faster to access.
Three SummarizedExperiment databases are provided as
HDF5-SummarizedExperiment files, including an unnormalized RGChannelSet
(red/green signals), an unnormalized MethylSet (methylated/unmethylated
signals) and a normalized GenomicRatioSet (DNAm fractions). For the latter,
DNAm fractions (logit2 Beta-values, or M-values) were normalized using the
out-of-band signal or “noob” method, an effective within-sample normalization
that removes signal artifacts [2].
Database files are stored as either HDF5 or HDF5-SummarizedExperiment. For
most R users, the latter files will be most convenient to work with. HDF5, or
hierarchical data format 5, combines compression and chunking for convenient
handling of large datasets. HDF5-SummarizedExperiment files combine the
benefits of HDF5 and SummarizedExperiment entities using a
DelayedArray-powered backend. Once an HDF5-SummarizedExperiment file is
loaded, it can be treated similarly to a SummarizedExperiment object in
active memory. That is, summary and subset operations execute rapidly, and
realization of large data chunks in active memory is delayed until called for
by the script (see examples).
Sample metadata are included with DNAm assays in the database files. Currently,
metadata variables include GEO record IDs for samples (GSM) and studies (GSE),
sample record titles, learned labels for tissue and disease, sample type
predictions from the MetaSRA-pipeline, and DNAm model-based predictions for
age, sex, and blood cell types. Access sample metadata from
SummarizedExperiment objects using the pData minfi function (see examples).
Examples in the data_analyses vignette illustrate some ways to utilize the
provided sample metadata.
Provided metadata derives from the GSE-specific SOFT files, which contain
experiment, sample, and platform metadata. Considerable efforts were made to
learn, harmonize, and predict metadata labels. Certain types of info lacking
in the recountmethylation metadata may be available in the SOFT files,
especially if it is sample non-specific (e.g. methods text, PubMed ID, etc.)
or redundant with DNAm-derived metrics (e.g. DNAm summaries, predicted sex,
etc.).
It is good practice to validate the harmonized metadata with original metadata records, especially where labels are ambiguous or there is insufficient information for a given query. GEO GSM and GSE records can be viewed from a browser, or SOFT files may be downloaded directly. Packages like GEOmetadb and GEOquery are also useful to query and summarize GEO metadata.
HDF5-SummarizedExperiment exampleThis example shows basic handling for HDF5-SummarizedExperiment (a.k.a.
“h5se”) files. For these files, the getdb function returns the loaded file.
Thanks to a DelayedArray backend, even full-sized h5se databases can be
treated as if they were fully loaded into active memory.
The test h5se dataset includes sample metadata and noob-normalized
DNAm fractions (Beta-values) for chromosome 22 probes for 2 samples.
Datasets can be downloaded using the getdb series of functions
(see ?getdb for details), where the dfp argument specifies the
download destination. The test h5se file is included in the package
“inst” directory, and can be loaded as follows.
dn <- "remethdb-h5se_gr-test_0-0-1_1590090412"
path <- system.file("extdata", dn, package = "recountmethylation")
h5se.test <- HDF5Array::loadHDF5SummarizedExperiment(path)Common characterization functions can be used on the dataset after it has been
loaded. These include functions for SummarizedExperiment-like objects, such
as the getBeta, pData, and getAnnotation minfi functions. First, inspect
the dataset using standard functions like class, dim, and summary as
follows.
class(h5se.test) # inspect object class## [1] "GenomicRatioSet"
## attr(,"package")
## [1] "minfi"dim(h5se.test) # get object dimensions## [1] 8552 2summary(h5se.test) # summarize dataset components## [1] "GenomicRatioSet object of length 8552 with 0 metadata columns"Access the sample metadata for the 2 available samples using pData.
h5se.md <- minfi::pData(h5se.test) # get sample metadata
dim(h5se.md) # get metadata dimensions## [1] 2 19colnames(h5se.md) # get metadata column names## [1] "gsm" "gsm_title" "gseid" "disease"
## [5] "tissue" "sampletype" "arrayid_full" "basename"
## [9] "age" "predage" "sex" "predsex"
## [13] "predcell.CD8T" "predcell.CD4T" "predcell.NK" "predcell.Bcell"
## [17] "predcell.Mono" "predcell.Gran" "storage"Next get CpG probe-specific DNAm fractions, or “Beta-values”, with getBeta
(rows are probes, columns are samples).
h5se.bm <- minfi::getBeta(h5se.test) # get dnam fractions
dim(h5se.bm) # get dnam fraction dimensions## [1] 8552 2colnames(h5se.bm) <- h5se.test$gsm # assign sample ids to dnam fractions
knitr::kable(head(h5se.bm), align = "c") # show table of dnam fractions | GSM1038308 | GSM1038309 | |
|---|---|---|
| cg00017461 | 0.9807283 | 0.9746836 |
| cg00077299 | 0.3476970 | 0.3456837 |
| cg00079563 | 0.8744652 | 0.9168005 |
| cg00087182 | 0.9763206 | 0.9760947 |
| cg00093544 | 0.0225112 | 0.0265087 |
| cg00101350 | 0.9736359 | 0.9789818 |
Access manifest information for probes with getAnnotation. This includes the
bead addresses, probe type, and genome coordinates and regions. For full details
about the probe annotations, consult the minfi and Illumina platform documentation.
an <- minfi::getAnnotation(h5se.test) # get platform annotation
dim(an) # get annotation dimensions## [1] 8552 33colnames(an) # get annotation column names## [1] "chr" "pos"
## [3] "strand" "Name"
## [5] "AddressA" "AddressB"
## [7] "ProbeSeqA" "ProbeSeqB"
## [9] "Type" "NextBase"
## [11] "Color" "Probe_rs"
## [13] "Probe_maf" "CpG_rs"
## [15] "CpG_maf" "SBE_rs"
## [17] "SBE_maf" "Islands_Name"
## [19] "Relation_to_Island" "Forward_Sequence"
## [21] "SourceSeq" "Random_Loci"
## [23] "Methyl27_Loci" "UCSC_RefGene_Name"
## [25] "UCSC_RefGene_Accession" "UCSC_RefGene_Group"
## [27] "Phantom" "DMR"
## [29] "Enhancer" "HMM_Island"
## [31] "Regulatory_Feature_Name" "Regulatory_Feature_Group"
## [33] "DHS"ant <- as.matrix(t(an[c(1:4), c(1:3, 5:6, 9, 19, 24, 26)])) # subset annotation
knitr::kable(ant, align = "c") # show annotation table| cg00017461 | cg00077299 | cg00079563 | cg00087182 | |
|---|---|---|---|---|
| chr | chr22 | chr22 | chr22 | chr22 |
| pos | 30663316 | 18632618 | 43253521 | 24302043 |
| strand | - | + | + | + |
| AddressA | 31616369 | 13618325 | 65630302 | 37797387 |
| AddressB | 70798487 | 37626331 | 55610348 | 20767312 |
| Type | I | I | I | I |
| Relation_to_Island | OpenSea | N_Shore | Island | N_Shore |
| UCSC_RefGene_Name | OSM | USP18 | ARFGAP3;ARFGAP3 | GSTT2B;GSTT2 |
| UCSC_RefGene_Group | TSS1500 | TSS200 | TSS200;TSS200 | Body;Body |
HDF5 database and exampleTo provide more workflow options, bead-specific red and green signal data have
been provided with sample metadata in an HDF5/h5 file. This example shows
how to handle objects of this type with recountmethylation.
The test h5 file includes metadata and bead-specific signals from
chromosome 22 for the same 2 samples as in the h5se test file.
Note getdb functions for h5 files simply return the database path.
Since the test h5 file has also been included in the package “inst” folder,
get the path to load the file as follows.
dn <- "remethdb-h5_rg-test_0-0-1_1590090412.h5" # get the h5se directory name
h5.test <- system.file("extdata", "h5test", dn,
package = "recountmethylation") # get the h5se dir pathUse the file path to read data into an RGChannelSet with the getrg
function. Setting all.gsm = TRUE obtains data for all samples in the
database files, while passing a vector of GSM IDs to gsmv argument
will query a subset of available samples. Signals from all available
probes are retrieved by default, and probe subsets can be obtained by
passing a vector of valid bead addresses to the cgv argument.
h5.rg <- getrg(dbn = h5.test, all.gsm = TRUE) # get red/grn signals from an h5 dbTo avoid exhausting active memory with the full-sized h5 dataset, provide
either gsmv or cgv to getrg, and set either all.cg or all.gsm to
FALSE (see ?getrg for details).
As in the previous example, use pData and getAnnotation to get sample
metadata and array manifest information, respectively. Access the green and
red signal matrices in the RGChannelSet with the getRed and getGreen
minfi functions.
h5.red <- minfi::getRed(h5.rg) # get red signal matrix
h5.green <- minfi::getGreen(h5.rg) # get grn signal matrix
dim(h5.red) # get dimensions of red signal matrix## [1] 11162 2knitr::kable(head(h5.red), align = "c") # show first rows of red signal matrix| GSM1038308 | GSM1038309 | |
|---|---|---|
| 10601475 | 1234 | 1603 |
| 10603366 | 342 | 344 |
| 10603418 | 768 | 963 |
| 10605304 | 2368 | 2407 |
| 10605460 | 3003 | 3322 |
| 10608343 | 357 | 399 |
knitr::kable(head(h5.green), align = "c") # show first rows of grn signal matrix| GSM1038308 | GSM1038309 | |
|---|---|---|
| 10601475 | 6732 | 8119 |
| 10603366 | 288 | 356 |
| 10603418 | 267 | 452 |
| 10605304 | 4136 | 4395 |
| 10605460 | 1395 | 1762 |
| 10608343 | 840 | 1269 |
identical(rownames(h5.red), rownames(h5.green)) # check cpg probe names identical## [1] TRUERows in these signal matrices map to bead addresses rather than probe IDs.
These matrices have more rows than the h5se test Beta-value matrix because
any type I probes use data from 2 beads each.
This section demonstrates validation using the test databases. Full code to reproduce this section is provided but not evaluated, as it involves a download from the GEO servers. As the disclaimer notes, it is good practice to validate data against the latest available GEO files. This step may be most useful for newer samples published close to the end compilation date (through November 7, 2020 for current version), which may be more prone to revisions at initial publication.
Use the gds_idat2rg function to download IDATs for the 2 test samples
and load these into a new RGChannelSet object. Do this by passing a vector
of GSM IDs to gsmv and the download destination to dfp. (note, chunks in
this section are fully executable, but not evaluated for this vignette).
# download from GEO
dlpath <- tempdir() # get a temp dir path
gsmv <- c("GSM1038308", "GSM1038309") # set sample ids to identify
geo.rg <- gds_idat2rg(gsmv, dfp = dlpath) # load sample idats into rgset
colnames(geo.rg) <- gsub("\\_.*", "", colnames(geo.rg)) # assign sample ids to columnsExtract the red and green signal matrices from geo.rg.
geo.red <- minfi::getRed(geo.rg) # get red signal matrix
geo.green <- minfi::getGreen(geo.rg) # get grn signal matrixMatch indices and labels between the GEO and h5 test signal matrices.
int.addr <- intersect(rownames(geo.red), rownames(h5.red)) # get probe address ids
geo.red <- geo.red[int.addr,] # subset geo rgset red signal
geo.green <- geo.green[int.addr,] # subset gro rgset grn signal
geo.red <- geo.red[order(match(rownames(geo.red), rownames(h5.red))),]
geo.green <- geo.green[order(match(rownames(geo.green), rownames(h5.green))),]
identical(rownames(geo.red), rownames(h5.red)) # check identical addresses, red
identical(rownames(geo.green), rownames(h5.green)) # check identical addresses, grn
class(h5.red) <- "integer"; class(h5.green) <- "integer" # set matrix data classes to integerFinally, compare the signal matrix data.
identical(geo.red, h5.red) # compare matrix signals, redidentical(geo.green, h5.green) # compare matrix signals, grnBefore comparing the GEO-downloaded data to data from the h5se.test database,
normalize the data using the same out-of-band or “noob” normalization technique
that was used to generate data in the h5se database.
geo.gr <- minfi::preprocessNoob(geo.rg) # get normalized se dataNext, extract the Beta-values.
geo.bm <- as.matrix(minfi::getBeta(geo.gr)) # get normalized dnam fractions matrixNow match row and column labels and indices.
h5se.bm <- as.matrix(h5se.bm) # set dnam fractions to matrix
int.cg <- intersect(rownames(geo.bm), rownames(h5se.bm))
geo.bm <- geo.bm[int.cg,] # subset fractions on shared probe ids
geo.bm <- geo.bm[order(match(rownames(geo.bm), rownames(h5se.bm))),]Finally, compare the two datasets.
identical(summary(geo.bm), summary(h5se.bm)) # check identical summary valuesidentical(rownames(geo.bm), rownames(h5se.bm)) # check identical probe idsThis section describes how to address potential issues with accessing the
database files or working with the DelayedArray based objects locally.
If repeated attempts to download the database compilation files fail, you may try the following:
First ensure your internet connection is stable and there is sufficient space at the download destination for the database file.
Second, try increasing your timeout duration beyond the default before
repeating the download attempt with getdb. Check the current timeout
for an R session with getOptions('timeout'), then manually increase
the timeout duration with options(timeout = new.time).
Finally, you may attempt to download a server file using command line
calls to your system terminal or console. For instance, on a Mac you
might try wget -r <file_url>. If this doesn’t work, you can again
attempt to increase the timeout duration and repeat the download attempt.
DelayedArray inputsUnexpected function behaviors may arise when using DelayedArray-based inputs.
These essentially arise from lacking interoperativity between normal matrices
and the DelayedArray-based matrices. Known examples include:
minfi::detectionP():Throws error for specific subsets of data, such as for queries of exactly 50 samples.
detectionP(rg[,1:50]) # get detection pvalues from rgset
"Error in .local(Red, Green, locusNames, controlIdx, TypeI.Red, TypeI.Green, dim(Red_grid) == dim(detP_sink_grid) are not all TRUE"minfi::preprocessFunnorm():Throws error when called for an RGChannelSet of type HDF5-SummarizedExperiment.
preprocessFunnorm(rg) # get noob-normalized data
"Error: 'preprocessFunnorm()' only supports matrix-backed minfi objects.""These and other related errors may be addressed by instantiating the data query,
or the data chunk, as a new non-DelayedArray object. For example, remake a
subset of the full h5se dataset, rg, as follows.
rg.h5se <- loadHDF5SummarizedExperiment(rg.path) # full h5se RGChannelSet
rg.sub <- rg.h5se[,c(1:20)] # subset samples of interest
rg.new <- RGChannelSet(Red = getRed(rg.sub),
Green = getGreen(rg.sub),
annotation = annotation(rg.sub)) # re-make as non-DA object
gr <- preprocessFunnorm(rg.new) # repeat preprocessingAlternatively, non-DelayedArray RGChannelSet objects can be readily generated from
the full h5 RGChannelSet database with the provided function getrg().
Consult the Data Analyses vignette and main manuscript for analysis examples and details about data compilations.
sessionInfo()## R version 4.1.2 (2021-11-01)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.4 LTS
##
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.14-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.14-bioc/R/lib/libRlapack.so
##
## 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
##
## attached base packages:
## [1] parallel stats4 stats graphics grDevices utils datasets
## [8] methods base
##
## other attached packages:
## [1] IlluminaHumanMethylation450kanno.ilmn12.hg19_0.6.0
## [2] ExperimentHub_2.2.1
## [3] AnnotationHub_3.2.2
## [4] BiocFileCache_2.2.1
## [5] dbplyr_2.1.1
## [6] reticulate_1.24
## [7] basilisk_1.6.0
## [8] HDF5Array_1.22.1
## [9] DelayedArray_0.20.0
## [10] Matrix_1.4-0
## [11] limma_3.50.1
## [12] gridExtra_2.3
## [13] ggplot2_3.3.5
## [14] knitr_1.37
## [15] recountmethylation_1.4.5
## [16] minfi_1.40.0
## [17] bumphunter_1.36.0
## [18] locfit_1.5-9.5
## [19] iterators_1.0.14
## [20] foreach_1.5.2
## [21] Biostrings_2.62.0
## [22] XVector_0.34.0
## [23] SummarizedExperiment_1.24.0
## [24] Biobase_2.54.0
## [25] MatrixGenerics_1.6.0
## [26] matrixStats_0.61.0
## [27] GenomicRanges_1.46.1
## [28] GenomeInfoDb_1.30.1
## [29] IRanges_2.28.0
## [30] S4Vectors_0.32.3
## [31] BiocGenerics_0.40.0
## [32] rhdf5_2.38.1
## [33] BiocStyle_2.22.0
##
## loaded via a namespace (and not attached):
## [1] plyr_1.8.6 splines_4.1.2
## [3] BiocParallel_1.28.3 digest_0.6.29
## [5] htmltools_0.5.2 magick_2.7.3
## [7] fansi_1.0.2 magrittr_2.0.2
## [9] memoise_2.0.1 tzdb_0.2.0
## [11] readr_2.1.2 annotate_1.72.0
## [13] askpass_1.1 siggenes_1.68.0
## [15] prettyunits_1.1.1 colorspace_2.0-3
## [17] blob_1.2.2 rappdirs_0.3.3
## [19] xfun_0.30 dplyr_1.0.8
## [21] crayon_1.5.0 RCurl_1.98-1.6
## [23] jsonlite_1.8.0 genefilter_1.76.0
## [25] GEOquery_2.62.2 survival_3.3-1
## [27] glue_1.6.2 gtable_0.3.0
## [29] zlibbioc_1.40.0 Rhdf5lib_1.16.0
## [31] scales_1.1.1 DBI_1.1.2
## [33] rngtools_1.5.2 Rcpp_1.0.8.2
## [35] xtable_1.8-4 progress_1.2.2
## [37] bit_4.0.4 mclust_5.4.9
## [39] preprocessCore_1.56.0 httr_1.4.2
## [41] dir.expiry_1.2.0 RColorBrewer_1.1-2
## [43] ellipsis_0.3.2 farver_2.1.0
## [45] pkgconfig_2.0.3 reshape_0.8.8
## [47] XML_3.99-0.9 sass_0.4.0
## [49] utf8_1.2.2 later_1.3.0
## [51] labeling_0.4.2 tidyselect_1.1.2
## [53] rlang_1.0.2 AnnotationDbi_1.56.2
## [55] BiocVersion_3.14.0 munsell_0.5.0
## [57] tools_4.1.2 cachem_1.0.6
## [59] cli_3.2.0 generics_0.1.2
## [61] RSQLite_2.2.10 evaluate_0.15
## [63] stringr_1.4.0 fastmap_1.1.0
## [65] yaml_2.3.5 bit64_4.0.5
## [67] beanplot_1.2 scrime_1.3.5
## [69] purrr_0.3.4 KEGGREST_1.34.0
## [71] nlme_3.1-155 doRNG_1.8.2
## [73] sparseMatrixStats_1.6.0 mime_0.12
## [75] nor1mix_1.3-0 xml2_1.3.3
## [77] biomaRt_2.50.3 compiler_4.1.2
## [79] interactiveDisplayBase_1.32.0 filelock_1.0.2
## [81] curl_4.3.2 png_0.1-7
## [83] tibble_3.1.6 bslib_0.3.1
## [85] stringi_1.7.6 basilisk.utils_1.6.0
## [87] highr_0.9 GenomicFeatures_1.46.5
## [89] lattice_0.20-45 multtest_2.50.0
## [91] vctrs_0.3.8 pillar_1.7.0
## [93] lifecycle_1.0.1 rhdf5filters_1.6.0
## [95] BiocManager_1.30.16 jquerylib_0.1.4
## [97] data.table_1.14.2 bitops_1.0-7
## [99] httpuv_1.6.5 rtracklayer_1.54.0
## [101] R6_2.5.1 BiocIO_1.4.0
## [103] promises_1.2.0.1 bookdown_0.24
## [105] codetools_0.2-18 MASS_7.3-55
## [107] assertthat_0.2.1 openssl_2.0.0
## [109] rjson_0.2.21 withr_2.5.0
## [111] GenomicAlignments_1.30.0 Rsamtools_2.10.0
## [113] GenomeInfoDbData_1.2.7 mgcv_1.8-39
## [115] hms_1.1.1 quadprog_1.5-8
## [117] grid_4.1.2 tidyr_1.2.0
## [119] base64_2.0 rmarkdown_2.13
## [121] DelayedMatrixStats_1.16.0 illuminaio_0.36.0
## [123] shiny_1.7.1 restfulr_0.0.131. Maden, S. K., Thompson, R. F., Hansen, K. D., and Nellore, A. (2020). Human methylome variation across Infinium 450K data on the Gene Expression Omnibus. bioRxiv. Available at: https://www.biorxiv.org/content/10.1101/2020.11.17.387548v2.
2. Triche, T. J., Weisenberger, D. J., Van Den Berg, D., Laird, P. W., and Siegmund, K. D. (2013). Low-level processing of Illumina Infinium DNA Methylation BeadArrays. Nucleic Acids Research 41, e90.
3. Sandoval, J., Heyn, H. A., Moran, S., Serra-Musach, J., Pujana, M. A., Bibikova, M., and Esteller, M. (2011). Validation of a DNA methylation microarry for 450,000 CpG sites in the human genome. Epigenetics 6, 692–702. Available at: http://www.landesbioscience.com/journals/epigenetics/article/16196/?nocache=1384341162.
4. Pidsley, R., Zotenko, E., Peters, T. J., Lawrence, M. G., Risbridger, G. P., Molloy, P., Van Djik, S., Muhlhausler, B., Stirzaker, C., and Clark, S. J. (2016). Critical evaluation of the Illumina MethylationEPIC BeadChip microarray for whole-genome DNA methylation profiling. Genome Biology 17. Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5055731/ [Accessed April 19, 2019].