---
title: "miaViz"
output:
BiocStyle::html_document:
fig_height: 7
fig_width: 10
toc: yes
toc_depth: 2
number_sections: true
vignette: >
%\VignetteIndexEntry{miaViz}
%\VignetteEngine{knitr::rmarkdown}
%\VignetteEncoding{UTF-8}
bibliography: references.bib
---
```{r, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#>"
)
knitr::opts_chunk$set(dev = "png", dev.args = list(type = "cairo-png"))
```
`miaViz` implements plotting function to work with `TreeSummarizedExperiment`
and related objects in a context of microbiome analysis. For more general
plotting function on `SummarizedExperiment` objects the `scater` package offers
several options, such as `plotColData`, `plotExpression` and `plotRowData`.
# Installation
To install `miaViz`, install `BiocManager` first, if it is not installed.
Afterwards use the `install` function from `BiocManager` and load `miaViz`.
```{r install, eval=FALSE}
if (!requireNamespace("BiocManager", quietly = TRUE))
install.packages("BiocManager")
BiocManager::install("miaViz")
```
```{r setup, message=FALSE}
library(miaViz)
library(scater)
data(GlobalPatterns, package = "mia")
```
# Abundance plotting
In contrast to other fields of sequencing based fields of research for which
expression of genes is usually studied, microbiome research uses the more
term ~Abundance~ to described the numeric data measured and analyzed.
Technically, especially in context of `SummarizedExperiment` objects, there is
no difference. Therefore `plotExpression` can be used to plot `Abundance` data of a particular feature.
```{r}
plotExpression(GlobalPatterns,
features = "549322", assay.type = "counts")
```
On the other hand, plotAbundance can be used to plot abundance by `rank`. A bar plot is returned showing the relative abundance within each sample for a given `rank`. At the same
time the `features` argument can be set to `NULL` (default).
```{r}
GlobalPatterns <- transformAssay(GlobalPatterns, method = "relabundance")
```
```{r}
plotAbundance(GlobalPatterns, rank = "Kingdom", assay.type = "relabundance")
```
If `rank` is set to null however then the bars will be colored by each
individual taxon. Please note that if you're doing this make sure to agglomerate
your data to a certain taxonomic hand before plotting.
```{r}
GlobalPatterns_king <- agglomerateByRank(GlobalPatterns, "Kingdom")
plotAbundance(GlobalPatterns_king, assay.type = "relabundance")
```
With subsetting to selected features the plot can be fine tuned.
```{r}
prev_phylum <- getPrevalent(GlobalPatterns, rank = "Phylum",
detection = 0.01, onRankOnly = TRUE)
```
```{r}
plotAbundance(GlobalPatterns[rowData(GlobalPatterns)$Phylum %in% prev_phylum],
rank = "Phylum",
assay.type = "relabundance")
```
The `features` argument is reused for plotting data along the different samples.
In the next example the ~SampleType~ is plotted along the samples. In this case
the result is a list, which can combined using external tools, for example
`patchwork`.
```{r}
library(patchwork)
plots <- plotAbundance(GlobalPatterns[rowData(GlobalPatterns)$Phylum %in% prev_phylum],
features = "SampleType",
rank = "Phylum",
assay.type = "relabundance")
plots$abundance / plots$SampleType +
plot_layout(heights = c(9, 1))
```
Further example about composition barplot can be found at Orchestrating Microbiome Analysis [@OMA].
# Prevalence plotting
To visualize prevalence within the dataset, two functions are available,
`plotFeaturePrevalence`, `plotPrevalenceAbundance` and `plotPrevalence`.
`plotFeaturePrevalence` produces a so-called landscape plot, which
visualizes the prevalence of samples across abundance thresholds.
```{r}
plotFeaturePrevalence(GlobalPatterns, rank = "Phylum",
detections = c(0, 0.001, 0.01, 0.1, 0.2))
```
`plotPrevalenceAbundance` plot the prevalence depending on the mean relative
abundance on the chosen taxonomic level.
```{r}
plotPrevalentAbundance(GlobalPatterns, rank = "Family",
colour_by = "Phylum") +
scale_x_log10()
```
`plotPrevalence` plot the number of samples and their prevalence across
different abundance thresholds. Abundance steps can be adjusted using the
`detections` argument, whereas the analyzed prevalence steps is set using the
`prevalences` argument.
```{r}
plotPrevalence(GlobalPatterns,
rank = "Phylum",
detections = c(0.01, 0.1, 1, 2, 5, 10, 20)/100,
prevalences = seq(0.1, 1, 0.1))
```
# Tree plotting
The information stored in the `rowTree` can be directly plotted. However,
sizes of stored trees have to be kept in mind and plotting of large trees
rarely makes sense.
For this example we limit the information plotted to the top 100 taxa as judged
by mean abundance on the genus level.
```{r, message=FALSE}
library(scater)
library(mia)
```
```{r}
altExp(GlobalPatterns,"Genus") <- agglomerateByRank(GlobalPatterns,"Genus")
altExp(GlobalPatterns,"Genus") <- addPerFeatureQC(altExp(GlobalPatterns,"Genus"))
rowData(altExp(GlobalPatterns,"Genus"))$log_mean <-
log(rowData(altExp(GlobalPatterns,"Genus"))$mean)
rowData(altExp(GlobalPatterns,"Genus"))$detected <-
rowData(altExp(GlobalPatterns,"Genus"))$detected / 100
top_taxa <- getTop(altExp(GlobalPatterns,"Genus"),
method="mean",
top=100L,
assay.type="counts")
```
Colour, size and shape of tree tips and nodes can be decorated based on data
present in the `SE` object or by providing additional information via the
`other_fields` argument. Note that currently information for nodes have to be
provided via the `other_fields` arguments.
Data will be matched via the `node` or `label` argument depending on which was
provided. `label` takes precedent.
```{r plot1, fig.cap="Tree plot using ggtree with tip labels decorated by mean abundance (colour) and prevalence (size)"}
plotRowTree(altExp(GlobalPatterns,"Genus")[top_taxa,],
tip_colour_by = "log_mean",
tip_size_by = "detected")
```
Tip and node labels can be shown as well. Setting `show_label = TRUE` shows the
tip labels only ...
```{r plot2, fig.cap="Tree plot using ggtree with tip labels decorated by mean abundance (colour) and prevalence (size). Tip labels of the tree are shown as well."}
plotRowTree(altExp(GlobalPatterns,"Genus")[top_taxa,],
tip_colour_by = "log_mean",
tip_size_by = "detected",
show_label = TRUE)
```
... whereas node labels can be selectively shown by providing a named logical
vector to `show_label`.
Please note that currently `ggtree` can only plot node labels in a rectangular
layout.
```{r plot3, fig.cap="Tree plot using ggtree with tip labels decorated by mean abundance (colour) and prevalence (size). Selected node and tip labels are shown."}
labels <- c("Genus:Providencia", "Genus:Morganella", "0.961.60")
plotRowTree(altExp(GlobalPatterns,"Genus")[top_taxa,],
tip_colour_by = "log_mean",
tip_size_by = "detected",
show_label = labels,
layout="rectangular")
```
Information can also be visualized on the edges of the tree plot.
```{r plot4, fig.cap="Tree plot using ggtree with tip labels decorated by mean abundance (colour) and edges labeled Kingdom (colour) and prevalence (size)"}
plotRowTree(altExp(GlobalPatterns,"Genus")[top_taxa,],
edge_colour_by = "Phylum",
tip_colour_by = "log_mean")
```
# Graph plotting
Similar to tree data, graph data can also be plotted in conjunction with
`SummarizedExperiment` objects. Since the graph data in itself cannot be stored
in a specialized slot, a graph object can be provided separately or as an
element from the `metedata`.
Here we load an example graph. As graph data, all objects types accepted by
`as_tbl_graph` from the `tidygraph` package are supported.
```{r}
data(col_graph)
```
In the following examples, the `weight` data is automatically generated from the
graph data. The `SummarizedExperiment` provided is required to have overlapping
rownames with the node names of the graph. Using this link the graph plot
can incorporate data from the `SummarizedExperiment`.
```{r}
plotColGraph(col_graph,
altExp(GlobalPatterns,"Genus"),
colour_by = "SampleType",
edge_colour_by = "weight",
edge_width_by = "weight",
show_label = TRUE)
```
As mentioned the graph data can be provided from the `metadata` of the
`SummarizedExperiment`.
```{r}
metadata(altExp(GlobalPatterns,"Genus"))$graph <- col_graph
```
This produces the same plot as shown above.
```{r include=FALSE}
plotColGraph(altExp(GlobalPatterns,"Genus"),
name = "graph",
colour_by = "SampleType",
edge_colour_by = "weight",
edge_width_by = "weight",
show_label = TRUE)
```
# Plotting of serial data
```{r, eval=FALSE}
if(!requireNamespace("miaTime", quietly = TRUE)){
remotes::install_github("microbiome/miaTime", upgrade = "never")
}
```
```{r, eval=FALSE}
# Load data from miaTime package
library("miaTime")
data(SilvermanAGutData, package="miaTime")
tse <- SilvermanAGutData
tse <- transformAssay(tse, method = "relabundance")
taxa <- getTop(tse, 2)
```
Data from samples collected along time can be visualized using `plotSeries`.
The `x` argument is used to reference data from the `colData` to use as
descriptor for ordering the data. The `y` argument selects the feature to show.
Since plotting a lot of features is not advised a maximum of 20 features can
plotted at the same time.
```{r, eval=FALSE}
plotSeries(tse,
x = "DAY_ORDER",
y = taxa,
colour_by = "Family")
```
If replicated data is present, data is automatically used for calculation of the
`mean` and `sd` and plotted as a range. Data from different assays can be used
for plotting via the `assay.type`.
```{r, eval=FALSE}
plotSeries(tse[taxa,],
x = "DAY_ORDER",
colour_by = "Family",
linetype_by = "Phylum",
assay.type = "relabundance")
```
Additional variables can be used to modify line type aesthetics.
```{r, eval=FALSE}
plotSeries(tse,
x = "DAY_ORDER",
y = getTop(tse, 5),
colour_by = "Family",
linetype_by = "Phylum",
assay.type = "counts")
```
# Plotting factor data
To visualize the relative relations between two groupings among the factor data,
two functions are available for the purpose; `plotColTile` and `plotRowTile`.
```{r}
data(GlobalPatterns, package="mia")
se <- GlobalPatterns
plotColTile(se,"SampleType","Primer") +
theme(axis.text.x.top = element_text(angle = 45, hjust = 0))
```
# DMN fit plotting
Searching for groups that are similar to each other among the samples, could be
approached with the Dirichlet Multinomial Mixtures [@DMM].
After using `runDMN` from the `mia` package, several k values as a number of clusters
are used to observe the best fit (see also `getDMN` and `getBestDMNFit`).
To visualize the fit using e.g. "laplace" as a measure of goodness of fit:
```{r}
data(dmn_se, package = "mia")
names(metadata(dmn_se))
# plot the fit
plotDMNFit(dmn_se, type = "laplace")
```
# Serial data ordination and trajectories
Principal Coordinates Analysis using Bray-Curtis dissimilarity on the
`hitchip1006` dataset:
```{r MDS, eval=FALSE}
library(miaTime)
data(hitchip1006, package = "miaTime")
tse <- hitchip1006
tse <- transformAssay(tse, method = "relabundance")
## Ordination with PCoA with Bray-Curtis dissimilarity
tse <- runMDS(tse, FUN = vegan::vegdist, method = "bray", name = "PCoA_BC",
assay.type = "relabundance", na.rm = TRUE)
# plot
p <- plotReducedDim(tse, dimred = "PCoA_BC")
p
```
Retrieving information about all available trajectories:
```{r, eval=FALSE}
library(dplyr)
# List subjects with two time points
selected.subjects <- names(which(table(tse$subject)==2))
# Subjects counts per number of time points available in the data
table(table(tse$subject)) %>% as.data.frame() %>%
rename(Timepoints=Var1, Subjects=Freq)
```
Lets look at all trajectories having two time points in the data:
```{r, eval=FALSE}
# plot
p + geom_path(aes(x=X1, y=X2, group=subject),
arrow=arrow(length = unit(0.1, "inches")),
# combining ordination data and metadata then selecting the subjects
# Note, scuttle::makePerCellDF could also be used for the purpose.
data = subset(data.frame(reducedDim(tse), colData(tse)),
subject %in% selected.subjects) %>% arrange(time))+
labs(title = "All trajectories with two time points")+
theme(plot.title = element_text(hjust = 0.5))
```
Filtering the two time point trajectories by divergence and displaying top 10%:
```{r, eval=FALSE}
library(miaTime)
# calculating step wise divergence based on the microbial profiles
tse <- getStepwiseDivergence(tse, group = "subject", time_field = "time")
# retrieving the top 10% divergent subjects having two time points
top.selected.subjects <- subset(data.frame(reducedDim(tse), colData(tse)),
subject %in% selected.subjects) %>%
top_frac(0.1, time_divergence) %>% select(subject) %>% .[[1]]
# plot
p + geom_path(aes(x=X1, y=X2,
color=time_divergence, group=subject),
# the data is sorted in descending order in terms of time
# since geom_path will use the first occurring observation
# to color the corresponding segment. Without the sorting
# geom_path will pick up NA values (corresponding to initial time
# points); breaking the example.
data = subset(data.frame(reducedDim(tse), colData(tse)),
subject %in% top.selected.subjects) %>%
arrange(desc(time)),
# arrow end is reversed, due to the earlier sorting.
arrow=arrow(length = unit(0.1, "inches"), ends = "first"))+
labs(title = "Top 10% divergent trajectories from time point one to two")+
scale_color_gradient2(low="white", high="red")+
theme(plot.title = element_text(hjust = 0.5))
```
Plotting an example of the trajectory with the maximum total divergence:
```{r, eval=FALSE}
# Get subject with the maximum total divergence
selected.subject <- data.frame(reducedDim(tse), colData(tse)) %>%
group_by(subject) %>%
summarise(total_divergence = sum(time_divergence, na.rm = TRUE)) %>%
filter(total_divergence==max(total_divergence)) %>% select(subject) %>% .[[1]]
# plot
p + geom_path(aes(x=X1, y=X2, group=subject),
data = subset(data.frame(reducedDim(tse), colData(tse)),
subject %in% selected.subject) %>% arrange(time),
arrow=arrow(length = unit(0.1, "inches")))+
labs(title = "Longest trajectory by divergence")+
theme(plot.title = element_text(hjust = 0.5))
```
More examples and materials are available at Orchestrating Microbiome Analysis [@OMA].
# Session info
```{r}
sessionInfo()
```
# References