Shinyngs is an R package designed to facilitate downstream analysis of RNA-seq and similar expression data with various exploratory plots and data mining tools. It is unrelated to the recently published Shiny Transcritome Analysis Resource Tool (START), though it was probably developed at the same time as that work.
A companion R package, zhangneurons, contains an example dataset to illustrate the features of Shinyngs, as well as the code required to produce it.
A Shinyngs example is running at https://pinin4fjords.shinyapps.io/shinyngs_example/ and contains a subset of the example data (due to limited resources on shinyapps.io).
Shinyngs differs to START and other similar applications (see also Degust), in that no effort is made to provide analysis capabilities. The envisaged process is:
- RNA-seq data is analysed, producing a set of matrices, and p/q values generated for a given set of comparisons.
- Matrix and comparison data is loaded into the modified SummarizedExperiment structure provided by Shinyngs, and serialised. This is easily automated.
- Serialised object used as input to autmoatically produce the Shiny app using Shinyngs.
There are a great many experimental designs and analysis methods, and in building Shinyngs I’ve taken the view that analysis is best left to the analyst. The envisaged use case is that of a bioinformatician attempting to convey results of analysis to non-experts.
ShinyNGS provides a number of capabilities you may not find in other applications:
- Simple selection of gene sets by name/ annotation to modify the plots and tables shown.
- Progressive filters for differential analysis: “Show me all genes differential in these contrasts but NOT in these other contrasts”
- Large variety of visualisations: row-wise clustering, UpSet-style intersection plots, gene set enrichment barcode plots etc.
- Allow rapid exploration of data output more or less straight from RNA-seq piplelines etc.
- Where more parameters are provided, extend the exploratory tools available - e.g. for differential expression.
- A variety of single and multiple-panel Shiny applications- currently heatmap, pca, boxplot, dendrogram, gene-wise barplot, various tables and an RNA-seq app combining all of these.
- Leveraging of libraries such as DataTables and Plotly for rich interactivity.
- Takes input in an extension of the commonly used
SummarizedExperimentformat, calledExploratorySummarizedExperiment - Interface kept simple where possible, with complexity automatically
added where required:
- Input field clutter reduced with the use of collapses from shinyBS (when installed).
- If a list of
ExploratorySummarizedExperiments is supplied (useful in situiations where the features are different beween matrices - e.g. from transcript- and gene- level analyses), a selection field will be provided. - If a selected experiment contains more than one assay, a selector will again be provided.
- For me: leveraging of Shiny modules. This makes re-using complex UI components much easier, and maintaining application code is orders of magnitude simpler as a result.
Shinyngs is built on Shiny ‘modules’- most of which are in single files
in the package code. As a consequence code is highly re-usable.
Documentation forthcoming, but take a look at how the selectmatrix
module is called by the PCA plots, boxplots etc.
shinyngs relies heavily on SummarizedExperiment. Formerly found in
the GenomicRanges package, it now has its own package on Bioconductor:
http://bioconductor.org/packages/release/bioc/html/SummarizedExperiment.html.
This requires a recent version of R.
Graphical enhancements are provided by shinyBS and shinyjs
Strong recommendation for Chrome over Firefox - everything renders much more nicely.
shinyngs is available as a Conda packge in Bioconda, as always it’s recommended to use a clean environment. With the Bioconda channel appropriately configured you can just do:
conda create -n shinyngs r-shinyngs
conda activate shinyngs(though I always recommend the mamba command in place of conda).
At the time of writing the dependency tree for arm64 was a bit
problematic. So just make and use Conda envs specifiying intel
architecture:
CONDA_SUBDIR=osx-64 conda create -n shinyngs r-shinyngs
conda activate shinyngs
conda config --env --set subdir osx-64Through the magic of the Bioconda and Biocontainers teams there is also a Docker image available.
devtools::install_github('pinin4fjords/shinyngs', upgrade_dependencies = FALSE)An example ExploratorySummarizedExperimentList based on the Zhang et
al study of neurons and glia
(http://www.jneurosci.org/content/34/36/11929.long) is available in a
separate package, and this can be used to demonstrate available
features.
Install the package like:
library(devtools)
install_github('pinin4fjords/zhangneurons')… and load and use the data like:
library(shinyngs)
library(zhangneurons)
data("zhangneurons")
app <- prepareApp("rnaseq", zhangneurons)
shiny::shinyApp(app$ui, app$server)The function eselistFromYAML() is provided to help build your own
objects given a config file.
A new feature (may be buggy) is the creation of Shiny apps from file complements:
make_app_from_files.R \
--assay_files raw.tsv,normalised_counts.tsv \
--sample_metadata samplesheet.csv \
--feature_metadata gene_meta.tsv \
--contrast_file contrasts.csv \
--differential_results treatment-saline-drug.deseq2.results.tsv \
--output_dir app \
--contrast_stats_assay 2 \
--unlog_foldchanges
(This script can be found under exec).
This is designed to take a regular file complement of
- Expression matrices
- Metadata (samples and features)
- Contrasts (which sample groups to compare)
- Differential resutls (e.g. from DESeq2) containing P values and fold changes
.. and produce an app.R. This currently covers the basic use cases and I haven’t go to the gene sets etc, that will be future work.
You can start the resulting app locally, by running the app.R
resulting from the above command.
See make_app_from_files.R --help for more info.
The following specified to make_app_from_files.R in addition to the
above will trigger a deployment to shinyapps.io where the app can be
viewed:
--deploy_app \
--shinyapps_account ACCOUNT \
--shinyapps_name APP_NAME
You must derive your token and secret from your shinyapps.io account and
set them in the environment variables SHINYAPPS_TOKEN and
SHINYAPPS_SECRET, respectively.
This is currently dependent on shinyngs having been installed via devtools, which doesn’t happen in the Conda install, but I’m trying to address that.
I’ve found it useful to reuse some of the plotting components in shinyngs to produce non-Shiny plot outputs for use in static reporting.
A generic complement of explortory plots can be generated like:
exploratory_plots.R \
--assay_files salmon.merged.gene_counts.tsv,normalised_counts.tsv,variance_stabilised_counts.tsv \
--assay_names raw,normalised,variance_stabilised \
--sample_metadata samplesheet.csv \
--contrast_variable treatment \
--outdir plots \
--feature_metadata gene_meta.tsv
See exploratory_plots.R --help for more info.
Differential analysis plots, currently just volcano plots, can be
generated with differential_plots.R. See exploratory_plots.R --help
for more info.
shinyngs has some good validation when building objects, to make sure that matrices are consistent with sample and feature annotations, and that the specified contrasts make sense. Accessing that logic by itself can be useful when writing FOM (feature/ observation matrix) workflows, so that is available separately like:
validate_fom_components.R \
--sample_metadata=testdata/samplesheet.csv \
--assay_files=testdata/SRP254919.salmon.merged.gene_counts.top1000cov.tsv \
--contrasts_file testdata/contrasts.csv \
--output_directory output
If --output_directory is specified, results are re-written (in a
consistent format, TSV by default) the specified location.
This script will error if there are inconsistencies between sample sheets, feature sets, matrices, and contrast specifications.
Technical information can be accessed via the package documentation:
?shinyngsMore user-oriented documentation and examples of how to build your own apps in the vignette.
This is also accessible via the vignette command:
vignette('shinyngs')- More useful non-RNAseq functionality to be added
I can be reached on @pinin4fjords with any queries. Other contributors welcome.