Differential gene expression (DGE) is one of the most common applications of RNA-sequencing (RNA-seq) data. This process allows for the elucidation of differentially expressed genes (DEGs) across two or more conditions. Interpretation of the DGE results can be non-intuitive and time consuming due to the variety of formats based on the tool of choice and the numerous pieces of information provided in these results files. Here we present an R package, ViDGER (Visualization of Differential Gene Expression Results using R), which contains nine functions that generate information-rich visualizations for the interpretation of DGE results from three widely-used tools, Cuffdiff, DESeq2, and edgeR.
The stable version of this package is available on Bioconductor. You can install it by running the following:
if (!requireNamespace("BiocManager", quietly=TRUE))
install.packages("BiocManager")
BiocManager::install("vidger")
The latest developmental version of ViDGER
can be installed via GitHub
using the devtools
package:
if (!require("devtools")) install.packages("devtools")
devtools::install_github("btmonier/vidger", ref = "devel")
Once installed, you will have access to the following functions:
vsBoxplot()
vsScatterPlot()
vsScatterMatrix()
vsDEGMatrix()
vsMAPlot()
vsMAMatrix()
vsVolcano()
vsVolcanoMatrix()
vsFourWay()
Further explanation will be given to how these functions work later on in the
documentation. For the following examples, three toy data sets will be used:
df.cuff
, df.deseq
, and df.edger
. Each of these data sets reflect the
three RNA-seq analyses this package covers. These can be loaded in the R
workspace by using the following command:
data(<data_set>)
Where <data_set>
is one of the previously mentioned data sets. Some of the
recurring elements that are found in each of these functions are the type
and d.factor
arguments. The type
argument tells the function how to
process the data for each analytical type (i.e. "cuffdiff"
, "deseq"
, or
"edger"
). The d.factor
argument is used specifically for DESeq2
objects
which we will discuss in the DESeq2 section. All other arguments are discussed
in further detail by looking at the respective help file for each functions
(i.e. ?vsScatterPlot
).
As mentioned earlier, three toy data sets are included with this package. In addition to these data sets, 5 “real-world” data sets were also used. All real-world data used is currently unpublished from ongoing collaborations. Summaries of this data can be found in the following tables:
Table 1: An overview of the toy data sets included in this package. In this table, each data set is summarized in terms of what analytical software was used, organism ID, experimental layout (replicates and treatments), number of transcripts (IDs), and size of the data object in terms of megabytes (MB).
Data | Software | Organism | Reps | Treat. | IDs | Size (MB) |
---|---|---|---|---|---|---|
df.cuff |
CuffDiff | H | 2 | 3 | 1200 | 0.2 |
sapiens | ||||||
df.deseq |
DESeq2 | D. | 2 | 3 | 29391 | 2.3 |
melanogaster | ||||||
df.deseq |
edgeR | A. | 2 | 3 | 724 | 0.1 |
thaliana |
Table 2: “Real-world” (RW) data set statistics. To test the reliability of our package, real data was used from human collections and several plant samples. Each data set is summarized in terms of organism ID, number of experimental samples (n), experimental conditions, and number of transcripts ( IDs).
Data | Organism | n | Exp. Conditions | IDs |
---|---|---|---|---|
RW-1 | H. | 10 | Two treatment dosages taken at two | 198002 |
sapiens | time points and one control sample | |||
taken at one time point | ||||
RW-2 | M. | 24 | Two phenotypes taken at four time | 63517 |
domestia | points (three replicates each) | |||
RW-3 | V. | 6 | Two conditions (three replicates | 59262 |
ripria: | each). | |||
bud | ||||
RW-4 | V. | 6 | Two conditions (three replicates | 17962 |
ripria: | each). | |||
shoot-tip | ||||
(7 days) | ||||
RW-5 | V. | 6 | Two conditions (three replicates | 19064 |
ripria: | each). | |||
shoot-tip | ||||
(21 days) |
Box plots are a useful way to determine the distribution of data. In this case
we can determine the distribution of FPKM or CPM values by using the
vsBoxPlot()
function. This function allows you to extract necessary
results-based data from analytical objects to create a box plot comparing
\(log_{10}\) (FPKM or CPM) distributions for experimental treatments.
vsBoxPlot(
data = df.cuff, d.factor = NULL, type = 'cuffdiff', title = TRUE,
legend = TRUE, grid = TRUE
)
Figure 1: A box plot example using the vsBoxPlot()
function with
cuffdiff
data. In this example, FPKM distributions for each treatment within
an experiment are shown in the form of a box and whisker plot.
vsBoxPlot(
data = df.deseq, d.factor = 'condition', type = 'deseq',
title = TRUE, legend = TRUE, grid = TRUE
)
Figure 2: A box plot example using the vsBoxPlot()
function with
DESeq2
data. In this example, FPKM distributions for each treatment within
an experiment are shown in the form of a box and whisker plot.
vsBoxPlot(
data = df.edger, d.factor = NULL, type = 'edger',
title = TRUE, legend = TRUE, grid = TRUE
)
Figure 3: A box plot example using the vsBoxPlot()
function with edgeR
data. In this example, CPM distributions for each treatment within an
experiment are shown in the form of a box and whisker plot
vsBoxPlot()
can allow for different iterations to showcase data
distribution. These changes can be implemented using the aes
parameter.
Currently, there are 6 different variants:
box
: standard box plotviolin
: violin plotboxdot
: box plot with dot plot overlayviodot
: violin plot with dot plot overlayviosumm
: violin plot with summary stats overlaynotch
: box plot with notchbox
variantdata("df.edger")
vsBoxPlot(
data = df.edger, d.factor = NULL, type = "edger", title = TRUE,
legend = TRUE, grid = TRUE, aes = "box"
)
Figure 4: A box plot example using the aes
parameter: box
violin
variantdata("df.edger")
vsBoxPlot(
data = df.edger, d.factor = NULL, type = "edger", title = TRUE,
legend = TRUE, grid = TRUE, aes = "violin"
)
Figure 5: A box plot example using the aes
parameter: violin
boxdot
variantdata("df.edger")
vsBoxPlot(
data = df.edger, d.factor = NULL, type = "edger", title = TRUE,
legend = TRUE, grid = TRUE, aes = "boxdot"
)
Figure 6: A box plot example using the aes
parameter: boxdot
viodot
variantdata("df.edger")
vsBoxPlot(
data = df.edger, d.factor = NULL, type = "edger", title = TRUE,
legend = TRUE, grid = TRUE, aes = "viodot"
)
Figure 7: A box plot example using the aes
parameter: viodot
viosumm
variantdata("df.edger")
vsBoxPlot(
data = df.edger, d.factor = NULL, type = "edger", title = TRUE,
legend = TRUE, grid = TRUE, aes = "viosumm"
)
Figure 8: A box plot example using the aes
parameter: viosumm
notch
variantdata("df.edger")
vsBoxPlot(
data = df.edger, d.factor = NULL, type = "edger", title = TRUE,
legend = TRUE, grid = TRUE, aes = "notch"
)
Figure 9: A box plot example using the aes
parameter: notch
In addition to aesthetic changes, the fill color of each variant can
also be changed. This can be implemented by modifiying the fill.color
parameter.
The palettes that can be used for this parameter are based off of the
palettes found in the RColorBrewer
package. A visual list of all the palettes can be found
here.
data("df.edger")
vsBoxPlot(
data = df.edger, d.factor = NULL, type = "edger", title = TRUE,
legend = TRUE, grid = TRUE, aes = "box", fill.color = "RdGy"
)
Figure 10: Color variant 1
A box plot example using the fill.color
parameter: RdGy
.
data("df.edger")
vsBoxPlot(
data = df.edger, d.factor = NULL, type = "edger", title = TRUE,
legend = TRUE, grid = TRUE, aes = "viosumm", fill.color = "Paired"
)
Figure 11: Color variant 2
A violin plot example using the fill.color
parameter: Paired
with the aes
parameter: viosumm
.
data("df.edger")
vsBoxPlot(
data = df.edger, d.factor = NULL, type = "edger", title = TRUE,
legend = TRUE, grid = TRUE, aes = "notch", fill.color = "Greys"
)
Figure 12: Color variant 3
A notched box plot example using the fill.color
parameter: Greys
with the aes
parameter: notch
. Using these parameters,
we can also generate grey-scale plots.
This example will look at a basic scatter plot function, vsScatterPlot()
.
This function allows you to visualize comparisons of \(log_{10}\) values of
either FPKM or CPM measurements of two treatments depending on analytical type.
vsScatterPlot(
x = 'hESC', y = 'iPS', data = df.cuff, type = 'cuffdiff',
d.factor = NULL, title = TRUE, grid = TRUE
)
Figure 13: A scatterplot example using the vsScatterPlot()
function with
Cuffdiff
data. In this visualization, \(log_{10}\) comparisons are made of
fragments per kilobase of transcript per million mapped reads (FPKM)
measurments. The dashed line represents regression line for the comparison.
vsScatterPlot(
x = 'treated_paired.end', y = 'untreated_paired.end',
data = df.deseq, type = 'deseq', d.factor = 'condition',
title = TRUE, grid = TRUE
)
Figure 14: A scatterplot example using the vsScatterPlot()
function with
DESeq2
data. In this visualization, \(log_{10}\) comparisons are made of
fragments per kilobase of transcript per million mapped reads (FPKM)
measurments. The dashed line represents regression line for the comparison.
vsScatterPlot(
x = 'WM', y = 'MM', data = df.edger, type = 'edger',
d.factor = NULL, title = TRUE, grid = TRUE
)
Figure 15: A scatterplot example using the vsScatterPlot()
function with
edgeR
data. In this visualization, \(log_{10}\) comparisons are made of
fragments per kilobase of transcript per million mapped reads (FPKM)
measurments. The dashed line represents regression line for the comparison.
This example will look at an extension of the vsScatterPlot()
function which
is vsScatterMatrix()
. This function will create a matrix of all possible
comparisons of treatments within an experiment with additional info.
vsScatterMatrix(
data = df.cuff, d.factor = NULL, type = 'cuffdiff',
comp = NULL, title = TRUE, grid = TRUE, man.title = NULL
)
Figure 16: A scatterplot matrix example using the vsScatterMatrix()
function with Cuffdiff
data. Similar to the scatterplot function, this
visualization allows for all comparisons to be made within an experiment. In
addition to the scatterplot visuals, FPKM distributions (histograms) and
correlation (Corr) values are generated.
vsScatterMatrix(
data = df.deseq, d.factor = 'condition', type = 'deseq',
comp = NULL, title = TRUE, grid = TRUE, man.title = NULL
)
Figure 17: A scatterplot matrix example using the vsScatterMatrix()
function with DESeq2
data. Similar to the scatterplot function, this
visualization allows for all comparisons to be made within an experiment. In
addition to the scatterplot visuals, FPKM distributions (histograms) and
correlation (Corr) values are generated.
vsScatterMatrix(
data = df.edger, d.factor = NULL, type = 'edger', comp = NULL,
title = TRUE, grid = TRUE, man.title = NULL
)
Figure 18: A scatterplot matrix example using the vsScatterMatrix()
function with edgeR
data. Similar to the scatterplot function, this
visualization allows for all comparisons to be made within an experiment. In
addition to the scatterplot visuals, FPKM distributions (histograms) and
correlation (Corr) values are generated.
Using the vsDEGMatrix()
function allows the user to visualize the number of
differentially expressed genes (DEGs) at a given adjusted p-value (padj =
) for each experimental treatment level. Higher color intensity correlates to
a higher number of DEGs.
vsDEGMatrix(
data = df.cuff, padj = 0.05, d.factor = NULL, type = 'cuffdiff',
title = TRUE, legend = TRUE, grid = TRUE
)
Figure 19: A matrix of differentially expressed genes (DEGs) at a given
p-value using the vsDEGMatrix()
function with Cuffdiff
data. With this
function, the user is able to visualize the number of DEGs ata given adjusted
p-value for each experimental treatment level. Higher color intensity
correlates to a higher number of DEGs.
vsDEGMatrix(
data = df.deseq, padj = 0.05, d.factor = 'condition',
type = 'deseq', title = TRUE, legend = TRUE, grid = TRUE
)
Figure 20: A matrix of differentially expressed genes (DEGs) at a given
p-value using the vsDEGMatrix()
function with DESeq2
data. With this
function, the user is able to visualize the number of DEGs ata given adjusted
p-value for each experimental treatment level. Higher color intensity
correlates to a higher number of DEGs.
vsDEGMatrix(
data = df.edger, padj = 0.05, d.factor = NULL, type = 'edger',
title = TRUE, legend = TRUE, grid = TRUE
)
Figure 21: A matrix of differentially expressed genes (DEGs) at a given
p-value using the vsDEGMatrix()
function with edgeR
data. With this
function, the user is able to visualize the number of DEGs ata given adjusted
p-value for each experimental treatment level. Higher color intensity
correlates to a higher number of DEGs.
A grey-scale option is available for this function if you wish to use a
grey-to-white gradient instead of the classic blue-to-white gradient. This
can be invoked by setting the grey.scale
parameter to TRUE
.
vsDEGMatrix(data = df.deseq, d.factor = "condition", type = "deseq",
grey.scale = TRUE
)
vsMAPlot()
visualizes the variance between two samples in terms of gene
expression values where logarithmic fold changes of count data are plotted
against mean counts. For more information on how each of the aesthetics are
plotted, please refer to the figure captions and Method S1.
vsMAPlot(
x = 'iPS', y = 'hESC', data = df.cuff, d.factor = NULL,
type = 'cuffdiff', padj = 0.05, y.lim = NULL, lfc = NULL,
title = TRUE, legend = TRUE, grid = TRUE
)
Figure 22: MA plot visualization using the vsMAPLot()
function with
Cuffdiff
data. LFCs are plotted mean counts to determine the variance
between two treatments in terms of gene expression. Blue nodes on the graph
represent statistically significant LFCs which are greater than a given value
than a user-defined LFC parameter. Green nodes indicate statistically
significant LFCs which are less than the user-defined LFC parameter. Gray
nodes are data points that are not statistically significant. Numerical values
in parantheses for each legend color indicate the number of transcripts that
meet the prior conditions. Triangular shapes represent values that exceed the
viewing area of the graph. Node size changes represent the magnitude of the
LFC values (i.e. larger shapes reflect larger LFC values). Dashed lines
indicate user-defined LFC values.
vsMAPlot(
x = 'treated_paired.end', y = 'untreated_paired.end',
data = df.deseq, d.factor = 'condition', type = 'deseq',
padj = 0.05, y.lim = NULL, lfc = NULL, title = TRUE,
legend = TRUE, grid = TRUE
)
Figure 23: MA plot visualization using the vsMAPLot()
function with
DESeq2
data. LFCs are plotted mean counts to determine the variance between
two treatments in terms of gene expression. Blue nodes on the graph represent
statistically significant LFCs which are greater than a given value than a
user-defined LFC parameter. Green nodes indicate statistically significant
LFCs which are less than the user-defined LFC parameter. Gray nodes are data
points that are not statistically significant. Numerical values in parantheses
for each legend color indicate the number of transcripts that meet the prior
conditions. Triangular shapes represent values that exceed the viewing area of
the graph. Node size changes represent the magnitude of the LFC values (i.e.
larger shapes reflect larger LFC values). Dashed lines indicate user-defined
LFC values.
vsMAPlot(
x = 'WW', y = 'MM', data = df.edger, d.factor = NULL,
type = 'edger', padj = 0.05, y.lim = NULL, lfc = NULL,
title = TRUE, legend = TRUE, grid = TRUE
)
Figure 24: MA plot visualization using the vsMAPLot()
function with
edgeR
data. LFCs are plotted mean counts to determine the variance between
two treatments in terms of gene expression. Blue nodes on the graph represent
statistically significant LFCs which are greater than a given value than a
user-defined LFC parameter. Green nodes indicate statistically significant
LFCs which are less than the user-defined LFC parameter. Gray nodes are data
points that are not statistically significant. Numerical values in parantheses
for each legend color indicate the number of transcripts that meet the prior
conditions. Triangular shapes represent values that exceed the viewing area of
the graph. Node size changes represent the magnitude of the LFC values (i.e.
larger shapes reflect larger LFC values). Dashed lines indicate user-defined
LFC values.
Similar to a scatter plot matrix, vsMAMatrix()
will produce visualizations
for all comparisons within your data set. For more information on how the
aesthetics are plotted in these visualizations, please refer to the figure
caption and Method S1.
vsMAMatrix(
data = df.cuff, d.factor = NULL, type = 'cuffdiff',
padj = 0.05, y.lim = NULL, lfc = 1, title = TRUE,
grid = TRUE, counts = TRUE, data.return = FALSE
)
Figure 25: A MA plot matrix using the vsMAMatrix()
function with Cuffdiff
data. Similar to the vsMAPlot()
function, vsMAMatrix()
will generate a
matrix of MA plots for all comparisons within an experiment. LFCs are plotted
mean counts to determine the variance between two treatments in terms of gene
expression. Blue nodes on the graph represent statistically significant LFCs
which are greater than a given value than a user-defined LFC parameter. Green
nodes indicate statistically significant LFCs which are less than the
user-defined LFC parameter. Gray nodes are data points that are not
statistically significant. Numerical values in parantheses for each legend
color indicate the number of transcripts that meet the prior conditions.
Triangular shapes represent values that exceed the viewing area of the graph.
Node size changes represent the magnitude of the LFC values (i.e. larger
shapes reflect larger LFC values). Dashed lines indicate user-defined LFC
values.
vsMAMatrix(
data = df.deseq, d.factor = 'condition', type = 'deseq',
padj = 0.05, y.lim = NULL, lfc = 1, title = TRUE,
grid = TRUE, counts = TRUE, data.return = FALSE
)
Figure 26: A MA plot matrix using the vsMAMatrix()
function with DESeq2
data. Similar to the vsMAPlot()
function, vsMAMatrix()
will generate a
matrix of MA plots for all comparisons within an experiment. LFCs are plotted
mean counts to determine the variance between two treatments in terms of gene
expression. Blue nodes on the graph represent statistically significant LFCs
which are greater than a given value than a user-defined LFC parameter. Green
nodes indicate statistically significant LFCs which are less than the
user-defined LFC parameter. Gray nodes are data points that are not
statistically significant. Numerical values in parantheses for each legend
color indicate the number of transcripts that meet the prior conditions.
Triangular shapes represent values that exceed the viewing area of the graph.
Node size changes represent the magnitude of the LFC values (i.e. larger
shapes reflect larger LFC values). Dashed lines indicate user-defined LFC
values.
vsMAMatrix(
data = df.edger, d.factor = NULL, type = 'edger',
padj = 0.05, y.lim = NULL, lfc = 1, title = TRUE,
grid = TRUE, counts = TRUE, data.return = FALSE
)
Figure 27: A MA plot matrix using the vsMAMatrix()
function with edgeR
data. Similar to the vsMAPlot()
function, vsMAMatrix()
will generate a
matrix of MA plots for all comparisons within an experiment. LFCs are plotted
mean counts to determine the variance between two treatments in terms of gene
expression. Blue nodes on the graph represent statistically significant LFCs
which are greater than a given value than a user-defined LFC parameter. Green
nodes indicate statistically significant LFCs which are less than the
user-defined LFC parameter. Gray nodes are data points that are not
statistically significant. Numerical values in parantheses for each legend
color indicate the number of transcripts that meet the prior conditions.
Triangular shapes represent values that exceed the viewing area of the graph.
Node size changes represent the magnitude of the LFC values (i.e. larger
shapes reflect larger LFC values). Dashed lines indicate user-defined LFC
values.
The next few visualizations will focus on ways to display differential gene
expression between two or more treatments. Volcano plots visualize the variance
between two samples in terms of gene expression values where the \(-log_{10}\) of
calculated p-values (y-axis) are a plotted against the \(log_2\) changes
(x-axis). These plots can be visualized with the vsVolcano()
function.
For more information on how each of the aesthetics are plotted, please refer
to the figure captions and Method S1.
vsVolcano(
x = 'iPS', y = 'hESC', data = df.cuff, d.factor = NULL,
type = 'cuffdiff', padj = 0.05, x.lim = NULL, lfc = NULL,
title = TRUE, legend = TRUE, grid = TRUE, data.return = FALSE
)
Figure 28: A volcano plot example using the vsVolcano()
function with
Cuffdiff
data. In this visualization, comparisons are made between the
\(-log_{10}\) p-value versus the \(log_2\) fold change (LFC) between two
treatments. Blue nodes on the graph represent statistically significant LFCs
which are greater than a given value than a user-defined LFC parameter. Green
nodes indicate statistically significant LFCs which are less than the
user-defined LFC parameter. Gray nodes are data points that are not
statistically significant. Numerical values in parantheses for each legend
color indicate the number of transcripts that meet the prior conditions. Left
and right brackets (< and >) represent values that exceed the viewing area of
the graph. Node size changes represent the magnitude of the LFC values (i.e.
larger shapes reflect larger LFC values). Vertical and horizontal lines
indicate user-defined LFC and adjusted p-values, respectively.
vsVolcano(
x = 'treated_paired.end', y = 'untreated_paired.end',
data = df.deseq, d.factor = 'condition', type = 'deseq',
padj = 0.05, x.lim = NULL, lfc = NULL, title = TRUE,
legend = TRUE, grid = TRUE, data.return = FALSE
)
Figure 29: A volcano plot example using the vsVolcano()
function with
DESeq2
data. In this visualization, comparisons are made between the
\(-log_{10}\) p-value versus the \(log_2\) fold change (LFC) between two
treatments. Blue nodes on the graph represent statistically significant LFCs
which are greater than a given value than a user-defined LFC parameter. Green
nodes indicate statistically significant LFCs which are less than the
user-defined LFC parameter. Gray nodes are data points that are not
statistically significant. Numerical values in parantheses for each legend
color indicate the number of transcripts that meet the prior conditions. Left
and right brackets (< and >) represent values that exceed the viewing area of
the graph. Node size changes represent the magnitude of the LFC values (i.e.
larger shapes reflect larger LFC values). Vertical and horizontal lines
indicate user-defined LFC and adjusted p-values, respectively.
vsVolcano(
x = 'WW', y = 'MM', data = df.edger, d.factor = NULL,
type = 'edger', padj = 0.05, x.lim = NULL, lfc = NULL,
title = TRUE, legend = TRUE, grid = TRUE, data.return = FALSE
)