An overview of the epistack package
5 avril 2023
Package
epistack 1.5.4
Contents
1 Introduction
Example of epistack outputs
The epistack package main objective is the visualizations of stacks
of genomic tracks (such as, but not restricted to, ChIP-seq or
DNA methyation data)
centered at genomic regions of interest. epistack needs three
different inputs:
- a genomic score objects, such as ChIP-seq coverage or DNA methylation values,
provided as a
GRanges(easily obtained frombigwigorbamfiles) - a list of feature of interest, such as peaks or transcription start sites,
provided as a
GRanges(easily obtained fromgtforbedfiles) - a score to sort the features, such as peak height, or gene expression value
Each inputs are then combined in a single RangedSummarizedExperiment
object use by epistack’s
ploting functions.
After introducing epistack’s plotting capacity, this document will present two use cases:
- plotting ChIP-seq signal at peaks, to investigate peak quality
- plotting DNA methylation and ChIP-seq signals at gene promoters sorted according to gene expression, to investigate the associations between an epigenetic mark and gene expression
2 Epistack visualisation
Epistack is a visualisation package. It uses a RangedSummarizedExperiment
object as input, with
matrices embeded as assays. We will discuss how to
build such input obects in the next section. For now on, we will focus on
the visualisation functions using the example dataset included in the package.
The dataset can be accessed with:
library(GenomicRanges)
library(SummarizedExperiment)
library(epistack)
data("stackepi")
stackepi
#> class: RangedSummarizedExperiment
#> dim: 693 51
#> metadata(0):
#> assays(1): DNAme
#> rownames(693): ENSSSCG00000016737 ENSSSCG00000036350 ...
#> ENSSSCG00000024209 ENSSSCG00000048227
#> rowData names(3): gene_id exp score
#> colnames(51): window_1 window_2 ... window_50 window_51
#> colData names(0):2.1 The plotEpisatck() function
This dataset can be visualised with the plotEpistack() function.
The first parameter is the input RangedSummarizedExperiment object.
The second (optionnal) parameter, assays specifies which assay(s)
should be displayed as heatmap(s). In the stackepi
dataset, only one track is present, with an assay names DNAme.
Note that it is possible to have several different tracks embeded in the
same RangedSummarizedExperiment object, as demonstarted in the next sections.
An aditional metric_col is used, to display score associated with each
anchor region, such as expression values or peak scores. Optionaly,
the metric_col can be transformed before ploting using the
metric_transfunc parameters.
plotEpistack(
stackepi,
assay = "DNAme", metric_col = "exp",
ylim = c(0, 1), zlim = c(0, 1),
x_labels = c("-2.5kb", "TSS", "+2.5kb"),
titles = "DNA methylation", legends = "%mCpG",
metric_title = "Expression", metric_label = "log10(TPM+1)",
metric_transfunc = function(x) log10(x+1)
)
If a bin column is present, it is used to generate one average profile per bin.
stackepi <- addBins(stackepi, nbins = 5)
plotEpistack(
stackepi,
assay = "DNAme", metric_col = "exp",
ylim = c(0, 1), zlim = c(0, 1),
x_labels = c("-2.5kb", "TSS", "+2.5kb"),
titles = "DNA methylation", legends = "%mCpG",
metric_title = "Expression", metric_label = "log10(TPM+1)",
metric_transfunc = function(x) log10(x+1)
)
Colours can be changed using dedicated parameters:
plotEpistack(
stackepi,
assay = "DNAme", metric_col = "exp",
ylim = c(0, 1), zlim = c(0, 1),
x_labels = c("-2.5kb", "TSS", "+2.5kb"),
titles = "DNA methylation", legends = "%mCpG",
metric_title = "Expression", metric_label = "log10(TPM+1)",
metric_transfunc = function(x) log10(x+1),
tints = "dodgerblue",
bin_palette = rainbow
)
Text size, and other graphical parameters, can be changed using cex inside of
plotEpistack(). Indeed, additional arguments will be passed internaly to
par() (see ?par for more details).
plotEpistack(
stackepi,
assay = "DNAme", metric_col = "exp",
ylim = c(0, 1), zlim = c(0, 1),
x_labels = c("-2.5kb", "TSS", "+2.5kb"),
titles = "DNA methylation", legends = "%mCpG",
metric_title = "Expression", metric_label = "log10(TPM+1)",
metric_transfunc = function(x) log10(x+1),
cex = 0.4, cex.main = 0.6
)
2.2 Custom panel arrangements
Each panel can be plotted individually using dedicated functions. For example:
plotAverageProfile(
stackepi,
ylim = c(0, 1),
assay = "DNAme",
x_labels = c("-2.5kb", "TSS", "+2.5kb"),
)
And:
plotStackProfile(
stackepi,
assay = "DNAme",
x_labels = c("-2.5kb", "TSS", "+2.5kb"),
palette = hcl.colors,
zlim = c(0, 1)
)
It is therefore possible to arrange panels as you whish, using the multipanel framework of your choice (layout, grid, patchwork, etc.).
layout(matrix(1:3, ncol = 1), heights = c(1.5, 3, 0.5))
old_par <- par(mar = c(2.5, 4, 0.6, 0.6))
plotAverageProfile(
stackepi, assay = "DNAme",
x_labels = c("-2.5kb", "TSS", "+2.5kb"), ylim = c(0, 1),
)
plotStackProfile(
stackepi, assay = "DNAme",
x_labels = c("-2.5kb", "TSS", "+2.5kb"), zlim = c(0, 1),
palette = hcl.colors
)
plotStackProfileLegend(
zlim = c(0, 1),
palette = hcl.colors
)
par(old_par)
layout(1)
3 Example 1: ChIP-seq coverage at peaks
3.1 data origin
In this part, we will use example ChIP-seq data from the 2016 CSAMA course “Basics of ChIP-seq data analysis” by Aleksandra Pekowska and Simon Anders. Data can be found in this github repository: github.com/Bioconductor/CSAMA2016/tree/master/lab-5-chipseq/EpigeneticsCSAMA/inst/bedfiles
There is no need to download the data as it can be remotely parsed.
The data consists of two H3K27ac ChIP-seq replicates, an input control,
and one list of peak for each replicates. It has been generated in
mouse Embryonic Stem cells and been subseted to have only data from chromosome 6
to allow fast vignette generation (but epistack can deal with whole genome
ChIP-seq data!).
path_reads <- c(
rep1 = "https://raw.githubusercontent.com/Bioconductor/CSAMA2016/master/lab-5-chipseq/EpigeneticsCSAMA/inst/bedfiles/H3K27ac_rep1_filtered_ucsc_chr6.bed",
rep2 = "https://raw.githubusercontent.com/Bioconductor/CSAMA2016/master/lab-5-chipseq/EpigeneticsCSAMA/inst/bedfiles/H3K27ac_rep2_filtered_ucsc_chr6.bed",
input = "https://raw.githubusercontent.com/Bioconductor/CSAMA2016/master/lab-5-chipseq/EpigeneticsCSAMA/inst/bedfiles/ES_input_filtered_ucsc_chr6.bed"
)
path_peaks <- c(
peak1 = "https://raw.githubusercontent.com/Bioconductor/CSAMA2016/master/lab-5-chipseq/EpigeneticsCSAMA/inst/bedfiles/Rep1_peaks_ucsc_chr6.bed",
peak2 = "https://raw.githubusercontent.com/Bioconductor/CSAMA2016/master/lab-5-chipseq/EpigeneticsCSAMA/inst/bedfiles/Rep2_peaks_ucsc_chr6.bed"
)3.2 data loading
We first read the peaks using rtracklayer:
library(rtracklayer)
peaks <- lapply(path_peaks, import)Peaks from each replicates can be merged using GenomicRanges
union() function (loaded with rtracklayer).
We then rescue peaks metadata, and compute a new mean_score that we use
to arrange our peak list.
merged_peaks <- GenomicRanges::union(peaks[[1]], peaks[[2]])
scores_rep1 <- double(length(merged_peaks))
scores_rep1[findOverlaps(peaks[[1]], merged_peaks, select = "first")] <- peaks[[1]]$score
scores_rep2 <- double(length(merged_peaks))
scores_rep2[findOverlaps(peaks[[2]], merged_peaks, select = "first")] <- peaks[[2]]$score
peak_type <- ifelse(
scores_rep1 != 0 & scores_rep2 != 0, "Both", ifelse(
scores_rep1 != 0, "Rep1 only", "Rep2 only"
)
)
mcols(merged_peaks) <- DataFrame(scores_rep1, scores_rep2, peak_type)
merged_peaks$mean_scores <- apply((mcols(merged_peaks)[, c("scores_rep1", "scores_rep2")]), 1, mean)
merged_peaks <- merged_peaks[order(merged_peaks$mean_scores, decreasing = TRUE), ]
rm(scores_rep1, scores_rep2, peak_type)
merged_peaks
#> GRanges object with 3534 ranges and 4 metadata columns:
#> seqnames ranges strand | scores_rep1 scores_rep2
#> <Rle> <IRanges> <Rle> | <numeric> <numeric>
#> [1] chr6 91638787-91646132 * | 3182.78 3203.04
#> [2] chr6 56746391-56748809 * | 3217.75 3100.00
#> [3] chr6 29296913-29298755 * | 3132.86 3100.00
#> [4] chr6 5445473-5448143 * | 3100.00 3100.00
#> [5] chr6 29996830-29999535 * | 3100.00 3100.00
#> ... ... ... ... . ... ...
#> [3530] chr6 94556781-94557433 * | 0.00 50.10
#> [3531] chr6 119142482-119143018 * | 50.09 0.00
#> [3532] chr6 100424986-100425605 * | 50.08 0.00
#> [3533] chr6 66963731-66964304 * | 0.00 50.03
#> [3534] chr6 96545103-96545640 * | 50.01 0.00
#> peak_type mean_scores
#> <character> <numeric>
#> [1] Both 3192.91
#> [2] Both 3158.88
#> [3] Both 3116.43
#> [4] Both 3100.00
#> [5] Both 3100.00
#> ... ... ...
#> [3530] Rep2 only 25.050
#> [3531] Rep1 only 25.045
#> [3532] Rep1 only 25.040
#> [3533] Rep2 only 25.015
#> [3534] Rep1 only 25.005
#> -------
#> seqinfo: 1 sequence from an unspecified genome; no seqlengthsWe then import the ChIP-seq reads. In the example datasets,
they are provided as .bed files, but for ChIP-seq we recommand
importing .bigwig coverage files. Bam files can also be imported using
GenomicAlignments::readGAlignment*().
reads <- lapply(path_reads, import)3.3 Generating coverage matrices
We generate coverage matrices with ChIP-seq coverage signal
summarized around each ChIP-seq peaks.
Several tools exists to generate such coverage matrices. We will demonstrate the
normalizeToMatrix() method from EnrichedHeatmap.
Other alternatives include Repitools’ featureScores(),
or seqplots’ getPlotSetArray().
We will focus on the regions around peak centers, extended from +/-5kb with
a window size of 250 bp. We keep track of the extent of our region of interest
in a xlab variable, and specify unique column names for each matrix.
Note: when using ChIP-seq data from a bigwig file, use the value_column parameter
of the normalizeToMatrix() function.
library(EnrichedHeatmap)
coverage_matrices <- lapply(
reads,
function(x) {
normalizeToMatrix(
x,
resize(merged_peaks, width = 1, fix = "center"),
extend = 5000, w = 250,
mean_mode = "coverage"
)
}
)
xlabs <- c("-5kb", "peak center", "+5kb")We then add the peak coordinates and each matrix to a
RangedSummarizedExperiment object.
merged_peaks <- SummarizedExperiment(
rowData = merged_peaks,
assays = coverage_matrices
)3.4 Plotting
The plotEpistack() function will use the merged_peaks object to generate a
complex representation of the ChIP-seq signals around
the genomic feature of interests (here ChIP-seq peaks).
plotEpistack(
merged_peaks,
assays = c("rep1", "rep2", "input"),
tints = c("dodgerblue", "firebrick1", "grey"),
titles = c("Rep1", "Rep2" , "Input"),
x_labels = xlabs,
zlim = c(0, 4), ylim = c(0, 4),
metric_col = "mean_scores", metric_title = "Peak score",
metric_label = "score"
)
If a bin column is present in the input GRanges object, it will
be used to annotate the figure and to generate one average profile per bin
in the lower panels. Here we use the peak_type as our bin column.
rowRanges(merged_peaks)$bin <- rowRanges(merged_peaks)$peak_type
plotEpistack(
merged_peaks,
assays = c("rep1", "rep2", "input"),
tints = c("dodgerblue", "firebrick1", "grey"),
titles = c("Rep1", "Rep2" , "Input"),
x_labels = xlabs,
zlim = c(0, 4), ylim = c(0, 4),
metric_col = "mean_scores", metric_title = "Peak score", metric_label = "score",
bin_palette = colorRampPalette(c("darkorchid1", "dodgerblue", "firebrick1")),
npix_height = 300
)
We can also sort on the bins first, and then on peak score. Epistack will
respect the order in the GRanges object.
merged_peaks <- merged_peaks[order(
rowRanges(merged_peaks)$bin, rowRanges(merged_peaks)$mean_scores,
decreasing = c(FALSE, TRUE), method = "radix"
), ]
plotEpistack(
merged_peaks,
patterns = c("rep1", "rep2", "input"),
tints = c("dodgerblue", "firebrick1", "grey"),
titles = c("Rep1", "Rep2" , "Input"),
x_labels = xlabs,
zlim = c(0, 4), ylim = c(0, 4),
metric_col = "mean_scores", metric_title = "Peak score", metric_label = "score",
bin_palette = colorRampPalette(c("darkorchid1", "dodgerblue", "firebrick1")),
npix_height = 300
)
4 Example 2: DNA methylation levels and ChIP-seq coverage at gene promoters sorted according to expression levels
4.1 Obtaining the TSS coordinates
In this part, we will plot the epigenetic signals at gene promoters, or more
precisely around gene Transcription Start Sites (TSS).
TSS coordinates can be obtained from various sources. One can access
the Ensembl annotations using
biomaRt, download a .gtf file and parse it using rtracklayer’s import(), or use AnnotationHub and
ensembldb. It is however important to work with the same genome
version has the one used to align the ChIP-seq reads.
For simplicity, we will use EnrichedHeatmap example data.
load(
system.file("extdata", "chr21_test_data.RData",
package = "EnrichedHeatmap"),
verbose = TRUE
)
#> Loading objects:
#> H3K4me3
#> cgi
#> genes
#> meth
#> rpkm
tss <- promoters(genes, upstream = 0, downstream = 1)
tss$gene_id <- names(tss)
tss
#> GRanges object with 720 ranges and 1 metadata column:
#> seqnames ranges strand | gene_id
#> <Rle> <IRanges> <Rle> | <character>
#> ENSG00000141956.9 chr21 43299591 - | ENSG00000141956.9
#> ENSG00000141959.12 chr21 45719934 + | ENSG00000141959.12
#> ENSG00000142149.4 chr21 33245628 + | ENSG00000142149.4
#> ENSG00000142156.10 chr21 47401651 + | ENSG00000142156.10
#> ENSG00000142166.8 chr21 34696734 + | ENSG00000142166.8
#> ... ... ... ... . ...
#> ENSG00000270533.1 chr21 10476061 - | ENSG00000270533.1
#> ENSG00000270652.1 chr21 38315567 + | ENSG00000270652.1
#> ENSG00000270835.1 chr21 39577700 - | ENSG00000270835.1
#> ENSG00000271308.1 chr21 11169720 + | ENSG00000271308.1
#> ENSG00000271486.1 chr21 20993009 + | ENSG00000271486.1
#> -------
#> seqinfo: 1 sequence from an unspecified genome; no seqlengths4.2 Adding expression data to the TSS coordinates
Epistack can work with any units and any scores, not limited to
expression data. Here we will use gene expression data from an RNA-seq
experiment, in RPKM units, as it is the data format available in
EnrichedHeatmap example dataset.
To join the expression data to the TSS coordinates, we will use an
epistack utility function addMetricAndArrangeGRanges():
expr <- data.frame(
gene_id = names(rpkm),
expr = rpkm
)
epidata <- addMetricAndArrangeGRanges(
tss, expr,
gr_key = "gene_id",
order_key = "gene_id", order_value = "expr"
)
epidata
#> GRanges object with 720 ranges and 2 metadata columns:
#> seqnames ranges strand | gene_id expr
#> <Rle> <IRanges> <Rle> | <character> <numeric>
#> ENSG00000198618.4 chr21 20230097 + | ENSG00000198618.4 461.0990
#> ENSG00000160307.5 chr21 48025121 - | ENSG00000160307.5 145.4498
#> ENSG00000142192.16 chr21 27543446 - | ENSG00000142192.16 137.3838
#> ENSG00000183255.7 chr21 46293752 - | ENSG00000183255.7 131.7230
#> ENSG00000142168.10 chr21 33031935 + | ENSG00000142168.10 97.5167
#> ... ... ... ... . ... ...
#> ENSG00000264063.1 chr21 9826203 + | ENSG00000264063.1 0
#> ENSG00000233767.1 chr21 47492162 - | ENSG00000233767.1 0
#> ENSG00000184029.5 chr21 39493454 - | ENSG00000184029.5 0
#> ENSG00000232118.1 chr21 30743547 - | ENSG00000232118.1 0
#> ENSG00000235965.1 chr21 20418629 + | ENSG00000235965.1 0
#> -------
#> seqinfo: 1 sequence from an unspecified genome; no seqlengthsWe create 5 bins of genes of equal sizes depending on the expression levels,
with epistack utility function addBins():
epidata <- addBins(epidata, nbins = 5)
epidata
#> GRanges object with 720 ranges and 3 metadata columns:
#> seqnames ranges strand | gene_id expr
#> <Rle> <IRanges> <Rle> | <character> <numeric>
#> ENSG00000198618.4 chr21 20230097 + | ENSG00000198618.4 461.0990
#> ENSG00000160307.5 chr21 48025121 - | ENSG00000160307.5 145.4498
#> ENSG00000142192.16 chr21 27543446 - | ENSG00000142192.16 137.3838
#> ENSG00000183255.7 chr21 46293752 - | ENSG00000183255.7 131.7230
#> ENSG00000142168.10 chr21 33031935 + | ENSG00000142168.10 97.5167
#> ... ... ... ... . ... ...
#> ENSG00000264063.1 chr21 9826203 + | ENSG00000264063.1 0
#> ENSG00000233767.1 chr21 47492162 - | ENSG00000233767.1 0
#> ENSG00000184029.5 chr21 39493454 - | ENSG00000184029.5 0
#> ENSG00000232118.1 chr21 30743547 - | ENSG00000232118.1 0
#> ENSG00000235965.1 chr21 20418629 + | ENSG00000235965.1 0
#> bin
#> <numeric>
#> ENSG00000198618.4 1
#> ENSG00000160307.5 1
#> ENSG00000142192.16 1
#> ENSG00000183255.7 1
#> ENSG00000142168.10 1
#> ... ...
#> ENSG00000264063.1 5
#> ENSG00000233767.1 5
#> ENSG00000184029.5 5
#> ENSG00000232118.1 5
#> ENSG00000235965.1 5
#> -------
#> seqinfo: 1 sequence from an unspecified genome; no seqlengths4.3 Extracting the epigenetic signals
As previously described, we use EnrichedHeatmap’s
normalizeToMatrix() function to extract the signals, rename the signal
colmun names, and add them to the epidata GRanges object:
methstack <- normalizeToMatrix(
meth, epidata, value_column = "meth",
extend = 5000, w = 250, mean_mode = "absolute"
)
h3k4me3stack <- normalizeToMatrix(
H3K4me3, epidata, value_column = "coverage",
extend = 5000, w = 250, mean_mode = "coverage"
)
epidata <- SummarizedExperiment(
rowData = epidata,
assays = list(DNAme = methstack, H3K4me3 = h3k4me3stack)
)4.4 Epistack plotting
The epidata GRanges object is now ready to plot:
plotEpistack(
epidata,
tints = c("dodgerblue", "orange"),
zlim = list(c(0, 1), c(0, 25)), ylim = list(c(0, 1), c(0, 50)),
x_labels = c("-5kb", "TSS", "+5kb"),
legends = c("%mCpG", "Coverage"),
metric_col = "expr", metric_title = "Gene expression",
metric_label = "log10(RPKM+1)",
metric_transfunc = function(x) log10(x + 1),
npix_height = 300
)
5 Citation
To cite {epistack}, please refers to its HAL entry:
Safia Saci, Guillaume Devailly. epistack: An R package to visualise stack profiles of epigenomic signals. 2021, ⟨hal-03401251v2⟩
6 sessionInfo()
sessionInfo()
#> R version 4.2.1 (2022-06-23)
#> Platform: x86_64-pc-linux-gnu (64-bit)
#> Running under: Ubuntu 22.04.2 LTS
#>
#> Matrix products: default
#> BLAS: /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.10.0
#> LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.10.0
#>
#> locale:
#> [1] LC_CTYPE=fr_FR.UTF-8 LC_NUMERIC=C
#> [3] LC_TIME=fr_FR.UTF-8 LC_COLLATE=fr_FR.UTF-8
#> [5] LC_MONETARY=fr_FR.UTF-8 LC_MESSAGES=fr_FR.UTF-8
#> [7] LC_PAPER=fr_FR.UTF-8 LC_NAME=C
#> [9] LC_ADDRESS=C LC_TELEPHONE=C
#> [11] LC_MEASUREMENT=fr_FR.UTF-8 LC_IDENTIFICATION=C
#>
#> attached base packages:
#> [1] grid stats4 stats graphics grDevices utils datasets
#> [8] methods base
#>
#> other attached packages:
#> [1] EnrichedHeatmap_1.26.0 ComplexHeatmap_2.12.1
#> [3] rtracklayer_1.56.1 epistack_1.5.4
#> [5] SummarizedExperiment_1.26.1 Biobase_2.56.0
#> [7] MatrixGenerics_1.8.1 matrixStats_0.63.0
#> [9] GenomicRanges_1.48.0 GenomeInfoDb_1.32.4
#> [11] IRanges_2.30.1 S4Vectors_0.34.0
#> [13] BiocGenerics_0.42.0 BiocStyle_2.24.0
#>
#> loaded via a namespace (and not attached):
#> [1] bitops_1.0-7 fs_1.5.2 usethis_2.1.6
#> [4] devtools_2.4.3 doParallel_1.0.17 RColorBrewer_1.1-3
#> [7] rprojroot_2.0.3 tools_4.2.1 bslib_0.4.0
#> [10] utf8_1.2.2 R6_2.5.1 colorspace_2.0-3
#> [13] GetoptLong_1.0.5 withr_2.5.0 prettyunits_1.1.1
#> [16] processx_3.7.0 compiler_4.2.1 cli_3.4.1
#> [19] DelayedArray_0.22.0 bookdown_0.30 sass_0.4.2
#> [22] callr_3.7.1 stringr_1.4.0 digest_0.6.29
#> [25] Rsamtools_2.12.0 rmarkdown_2.14 XVector_0.36.0
#> [28] pkgconfig_2.0.3 htmltools_0.5.2 sessioninfo_1.2.2
#> [31] plotrix_3.8-2 fastmap_1.1.0 highr_0.9
#> [34] rlang_1.0.6 GlobalOptions_0.1.2 rstudioapi_0.13
#> [37] shape_1.4.6 jquerylib_0.1.4 BiocIO_1.6.0
#> [40] jsonlite_1.8.0 BiocParallel_1.30.4 RCurl_1.98-1.9
#> [43] magrittr_2.0.3 GenomeInfoDbData_1.2.8 Matrix_1.5-3
#> [46] Rcpp_1.0.9 fansi_1.0.3 lifecycle_1.0.3
#> [49] stringi_1.7.8 yaml_2.3.5 zlibbioc_1.42.0
#> [52] pkgbuild_1.3.1 parallel_4.2.1 crayon_1.5.1
#> [55] lattice_0.20-45 Biostrings_2.64.1 circlize_0.4.15
#> [58] locfit_1.5-9.6 magick_2.7.3 knitr_1.39
#> [61] ps_1.7.1 pillar_1.7.0 rjson_0.2.21
#> [64] codetools_0.2-18 pkgload_1.3.0 XML_3.99-0.12
#> [67] glue_1.6.2 evaluate_0.15 remotes_2.4.2
#> [70] BiocManager_1.30.19 vctrs_0.4.1 png_0.1-7
#> [73] foreach_1.5.2 purrr_0.3.4 clue_0.3-63
#> [76] cachem_1.0.6 xfun_0.38 restfulr_0.0.15
#> [79] tibble_3.1.7 iterators_1.0.14 GenomicAlignments_1.32.1
#> [82] memoise_2.0.1 cluster_2.1.3 ellipsis_0.3.2