R/run_app.R

In LieberInstitute/spatialLIBD: spatialLIBD: an R/Bioconductor package to visualize spatially-resolved transcriptomics data

#' Run the spatialLIBD Shiny Application
#'
#' This function runs the shiny application that allows users to interact
#' with the Visium spatial transcriptomics data from LIBD (by default) or
#' any other data that you have shaped according to our object structure.
#'
#' If you don't have the pseudo-bulked analysis results like we computed them
#' in our project <https://doi.org/10.1038/s41593-020-00787-0> you can
#' set `sce_layer`, `modeling_results` and `sig_genes` to `NULL`. Doing so
#' will disable the pseudo-bulked portion of the web application. See the
#' examples for one such case as well as the vignette that describes how
#' you can use `spatialLIBD` with public data sets provided by 10x Genomics.
#' That vignette is available at
#' <http://research.libd.org/spatialLIBD/articles/TenX_data_download.html>.
#'
#' @param spe Defaults to the output of
#' `fetch_data(type = 'spe')`. This is a
#' [SpatialExperiment-class][SpatialExperiment::SpatialExperiment-class]
#' object with the spot-level Visium data and information required for
#' visualizing the histology. See [fetch_data()] for more details.
#' @inheritParams sig_genes_extract
#' @param sig_genes The output of [sig_genes_extract_all()] which is a table
#' in long format with the modeling results. You can subset this if the object
#' requires too much memory.
#' @param docs_path A `character(1)` specifying the path to the directory
#' containing the website documentation files. The directory has to contain
#' the files: `documentation_sce_layer.md`, `documentation_spe.md`,
#' `favicon.ico`, `footer.html` and `README.md`.
#' @param title A character(1) specifying the title for the app.
#' @param spe_discrete_vars A `character()` vector of discrete variables that
#' will be available to visualize in the app. Basically, the set of variables
#' with spot-level groups. They will have to be present in `colData(spe)`.
#' @param spe_continuous_vars A `character()` vector of continuous variables
#' that will be available to visualize in the app using the same scale
#' as genes. They will have to be present in `colData(sce)`.
#' @param default_cluster A `character(1)` with the name of the main cluster
#' (discrete) variable to use. It will have to be present in both `colData(spe)`
#' and `colData(sce_layer)`.
#' @param auto_crop_default A `logical(1)` specifying the default value for
#' automatically cropping the images. Set this to `FALSE` if your images do not
#' follow the Visium grid size expectations, which are key for enabling
#' auto-cropping.
#' @param ... Other arguments passed to the list of golem options for running
#' the application.
#'
#' @export
#' @importFrom shiny shinyApp
#' @importFrom golem with_golem_options
#' @return A [shiny.appobj][shiny::shiny.appobj] that contains the input data.
#'
#' @examples
#' \dontrun{
#' ## The default arguments will download the data from the web
#' ## using fetch_data(). If this is the first time you have run this,
#' ## the files will need to be cached by ExperimentHub. Otherwise it
#' ## will re-use the files you have previously downloaded.
#' if (enough_ram(4e9)) {
#'     ## Obtain the necessary data
#'     if (!exists("spe")) spe <- fetch_data("spe")
#'
#'     ## Create the interactive website
#'     run_app(spe)
#'
#'     ## You can also run a custom version without the pseudo-bulked
#'     ## layer information. This is useful if you are only interested
#'     ## in the spatial transcriptomics features.
#'     run_app(spe,
#'         sce_layer = NULL, modeling_results = NULL, sig_genes = NULL,
#'         title = "spatialLIBD without layer info"
#'     )
#'
#'     ## When using shinyapps.io aim for less than 3 GB of RAM with your
#'     ## objects. Check each input object with:
#'     ## lobstr::obj_size(x)
#'     ## Do not create the large input objects on the app.R script before
#'     ## subsetting them. Do this outside app.R since the app.R script is
#'     ## run at shinyapps.io, so subsetting on that script to reduce the
#'     ## memory load is pointless. You have to do it outside of app.R.
#' }
#'
#' ## How to run locally the spatialDLPFC Sp09 spatialLIBD app. That is,
#' ## from http://research.libd.org/spatialDLPFC/#interactive-websites
#' ## how to run https://libd.shinyapps.io/spatialDLPFC_Visium_Sp09 locally.
#' if (enough_ram(9e9)) {
#'     ## Download the 3 main objects needed
#'     spe <- fetch_data("spatialDLPFC_Visium")
#'     sce_pseudo <- fetch_data("spatialDLPFC_Visium_pseudobulk")
#'     modeling_results <- fetch_data("spatialDLPFC_Visium_modeling_results")
#'
#'     ## These are optional commands to further reduce the memory required.
#'     #
#'     ## Keep only the "lowres" images. Reduces the object from 6.97 GB to 4.59 GB
#'     # imgData(spe) <- imgData(spe)[imgData(spe)$image_id == "lowres", ]
#'     ## Drop the regular counts (keep only the logcounts). Reduces the object
#'     ## from 4.59 GB to 2.45 GB.
#'     # counts(spe) <- NULL
#'
#'     ## For sig_genes_extract_all() to work
#'     sce_pseudo$spatialLIBD <- sce_pseudo$BayesSpace
#'     ## Compute the significant genes
#'     sig_genes <- sig_genes_extract_all(
#'         n = nrow(sce_pseudo),
#'         modeling_results = modeling_results,
#'         sce_layer = sce_pseudo
#'     )
#'     ## Reduce the memory from 423.73 MB to 78.88 MB
#'     lobstr::obj_size(sig_genes)
#'     sig_genes$in_rows <- NULL
#'     sig_genes$in_rows_top20 <- NULL
#'     lobstr::obj_size(sig_genes)
#'
#'     ## Specify the default variable
#'     spe$BayesSpace <- spe$BayesSpace_harmony_09
#'     ## Get all variables
#'     vars <- colnames(colData(spe))
#'
#'     ## Set default cluster colors
#'     colors_BayesSpace <- Polychrome::palette36.colors(28)
#'     names(colors_BayesSpace) <- c(1:28)
#'     m <- match(as.character(spe$BayesSpace_harmony_09), names(colors_BayesSpace))
#'     stopifnot(all(!is.na(m)))
#'     spe$BayesSpace_colors <- spe$BayesSpace_harmony_09_colors <- colors_BayesSpace[m]
#'
#'     ## Download documentation files we use
#'     temp_www <- file.path(tempdir(), "www")
#'     dir.create(temp_www)
#'     download.file(
#'         "https://raw.githubusercontent.com/LieberInstitute/spatialDLPFC/main/README.md",
#'         file.path(temp_www, "README.md")
#'     )
#'     download.file(
#'         "https://raw.githubusercontent.com/LieberInstitute/spatialDLPFC/main/code/deploy_app_k09/www/documentation_sce_layer.md",
#'         file.path(temp_www, "documentation_sce_layer.md")
#'     )
#'     download.file(
#'         "https://raw.githubusercontent.com/LieberInstitute/spatialDLPFC/main/code/deploy_app_k09/www/documentation_spe.md",
#'         file.path(temp_www, "documentation_spe.md")
#'     )
#'     download.file(
#'         "https://raw.githubusercontent.com/LieberInstitute/spatialDLPFC/main/img/favicon.ico",
#'         file.path(temp_www, "favicon.ico")
#'     )
#'     download.file(
#'         "https://raw.githubusercontent.com/LieberInstitute/spatialDLPFC/main/code/deploy_app_k09/www/footer.html",
#'         file.path(temp_www, "footer.html")
#'     )
#'     list.files(temp_www)
#'
#'     ## Run the app locally
#'     run_app(
#'         spe,
#'         sce_layer = sce_pseudo,
#'         modeling_results = modeling_results,
#'         sig_genes = sig_genes,
#'         title = "spatialDLPFC, Visium, Sp09",
#'         spe_discrete_vars = c( # this is the variables for the spe object not the sce_pseudo object
#'             "BayesSpace",
#'             "ManualAnnotation",
#'             vars[grep("^SpaceRanger_|^scran_", vars)],
#'             vars[grep("^BayesSpace_harmony", vars)],
#'             vars[grep("^BayesSpace_pca", vars)],
#'             "graph_based_PCA_within",
#'             "PCA_SNN_k10_k7",
#'             "Harmony_SNN_k10_k7",
#'             "manual_layer_label",
#'             "wrinkle_type",
#'             "BayesSpace_colors"
#'         ),
#'         spe_continuous_vars = c(
#'             "sum_umi",
#'             "sum_gene",
#'             "expr_chrM",
#'             "expr_chrM_ratio",
#'             vars[grep("^VistoSeg_", vars)],
#'             vars[grep("^layer_", vars)],
#'             vars[grep("^broad_", vars)]
#'         ),
#'         default_cluster = "BayesSpace",
#'         docs_path = temp_www
#'     )
#' }
#' ## See also:
#' ## * https://github.com/LieberInstitute/spatialDLPFC/tree/main/code/deploy_app_k09
#' ## * https://github.com/LieberInstitute/spatialDLPFC/tree/main/code/deploy_app_k09_position
#' ## * https://github.com/LieberInstitute/spatialDLPFC/tree/main/code/deploy_app_k09_position_noWM
#' ## * https://github.com/LieberInstitute/spatialDLPFC/tree/main/code/deploy_app_k16
#' ## * https://github.com/LieberInstitute/spatialDLPFC/tree/main/code/analysis_IF/03_spatialLIBD_app
#' }
run_app <- function(
        spe = fetch_data(type = "spe"),
        sce_layer = fetch_data(type = "sce_layer"),
        modeling_results = fetch_data(type = "modeling_results"),
        sig_genes = sig_genes_extract_all(
            n = nrow(sce_layer),
            modeling_results = modeling_results,
            sce_layer = sce_layer
        ),
        docs_path = system.file("app", "www", package = "spatialLIBD"),
        title = "spatialLIBD",
        spe_discrete_vars = c(
            "spatialLIBD",
            "GraphBased",
            "ManualAnnotation",
            "Maynard",
            "Martinowich",
            paste0("SNN_k50_k", 4:28),
            "SpatialDE_PCA",
            "SpatialDE_pool_PCA",
            "HVG_PCA",
            "pseudobulk_PCA",
            "markers_PCA",
            "SpatialDE_UMAP",
            "SpatialDE_pool_UMAP",
            "HVG_UMAP",
            "pseudobulk_UMAP",
            "markers_UMAP",
            "SpatialDE_PCA_spatial",
            "SpatialDE_pool_PCA_spatial",
            "HVG_PCA_spatial",
            "pseudobulk_PCA_spatial",
            "markers_PCA_spatial",
            "SpatialDE_UMAP_spatial",
            "SpatialDE_pool_UMAP_spatial",
            "HVG_UMAP_spatial",
            "pseudobulk_UMAP_spatial",
            "markers_UMAP_spatial"
        ),
        spe_continuous_vars = c(
            "cell_count",
            "sum_umi",
            "sum_gene",
            "expr_chrM",
            "expr_chrM_ratio"
        ),
        default_cluster = "spatialLIBD",
        auto_crop_default = TRUE,
        ...) {
    ## Run the checks in the relevant order
    stopifnot(length(default_cluster) == 1)
    stopifnot(default_cluster %in% spe_discrete_vars)

    spe <-
        check_spe(spe,
            variables = c(spe_discrete_vars, spe_continuous_vars)
        )

    ## Check sce_layer and modeling_results if needed
    if (!is.null(sce_layer)) {
        sce_layer <-
            check_sce_layer(sce_layer, variables = default_cluster)
        modeling_results <- check_modeling_results(modeling_results)
        ## No need to check sig_genes since sig_genes_extract_all() will fail
    }

    ## Check that the required documentation files exist
    stopifnot(all(file.exists(file.path(
        docs_path,
        c(
            "documentation_spe.md",
            "favicon.ico",
            "footer.html",
            "README.md"
        )
    ))))
    if (!is.null(sce_layer)) {
        ## Check required files when sce_layer is present
        stopifnot(file.exists(file.path(
            docs_path, "documentation_sce_layer.md"
        )))
    }

    with_golem_options(
        app = shinyApp(ui = app_ui, server = app_server),
        golem_opts = list(
            spe = spe,
            sce_layer = sce_layer,
            modeling_results = modeling_results,
            sig_genes = sig_genes,
            docs_path = docs_path,
            title = title,
            spe_discrete_vars = spe_discrete_vars,
            spe_continuous_vars = spe_continuous_vars,
            default_cluster = default_cluster,
            auto_crop_default = auto_crop_default,
            ...
        )
    )
}