R/ivyGlimpse.R

In ivygapSE: A SummarizedExperiment for Ivy-GAP data

#' simple app to explore image property quantifications in relation to survival and expression
#' @import shiny
#' @importFrom UpSetR upset
#' @importFrom utils packageVersion
#' @importFrom S4Vectors metadata
#' @import survminer
#' @import hwriter
#' @import SummarizedExperiment
#' @import survival
#' @import plotly
#' @importFrom graphics boxplot par
#' @importFrom stats median
#' @importFrom utils browseURL data
#' @rawNamespace import(ggplot2, except=last_plot)
#' @return Side effect of starting the app only.
#' @examples
#' if (interactive()) print(ivyGlimpse())
#' @export
ivyGlimpse = function() {

## START UI
# define gene sets from cbioPortal

    someSets = makeGeneSets()
    
    #ivySE = NULL
    #data(ivySE)
    load(system.file("data/ivySE.rda", package="ivygapSE"))
    
    sb = metadata(ivySE)$subBlockDetails
    sb = sb[!is.na(sb$survival_days),]
    
    feats = colnames(sb)
    
    ui = fluidPage(
     sidebarLayout(
      sidebarPanel(
       fluidRow(
          helpText("IvyGAP explorer: expression, clinical, and image-based data for glioblastoma patients; see background panel for more details")
       ),
       fluidRow(
          helpText("subBlockDetail features for selectables scatterplot")
       ),
       fluidRow(
        selectInput("x", "x", choices=feats, selected="normalized_area_it")
       ),
       fluidRow(
        selectInput("y", "y", choices=feats, selected="normalized_area_ct")
       ), 
       fluidRow(
        selectInput("gs", "cbioP sets", choices=names(someSets), selected=names(someSets)[1])
       ),
       fluidRow(
        helpText("Supported by NCI ITCR U01 CA214846 and U24 CA180996")
       ),
       width=3
      ),
      mainPanel(
       tabsetPanel(
        tabPanel("basic",
         fluidRow(
          column( 
           fluidRow( helpText("hover over for tumor donor ID; partition data by dragging over points to select; click on a specific point to visit IvyGAP clinical specimen page for that sample") ),
           fluidRow( plotlyOutput("xyplot") ), width=5 ),
          column( 
           fluidRow( helpText("Kaplan-Meier, grp=1 for selected donors") ),
           fluidRow( plotOutput("plot2") ), width=5)
          ),
         fluidRow(
          column( plotOutput("boxes1"), width=5 ),
          column( plotOutput("boxes2"), width=5 )
          )
         ),
        tabPanel("vocabulary", tableOutput("vocab")),
        tabPanel("background", 
         fluidRow(
           textOutput("versinfo"),
           textOutput("commentary"),
           plotOutput("upset")
           ),
         fluidRow(
           helpText("Read the ", a(href='http://help.brain-map.org/display/glioblastoma/Documentation?preview=/8028197/8454231/IvyOverview.pdf', 'Allen Institute Technical White Paper'))
          ),
         fluidRow(helpText("The README from the ", a(href='http://glioblastoma.alleninstitute.org/api/v2/well_known_file_download/305873915',"zip archive of expression data"))),
         fluidRow(verbatimTextOutput("bkgrd"))
         )
        )
       )
      )
     )
    
    ## END UI
    ## START SERVER
    
    featanno = c("normalized_area_le"   =   
    "Leading Edge (LE)",
    "normalized_area_lehbv" =
    "Hyperplastic blood vessels in leading edge (LEhbv)",
    "normalized_area_it"   =   
    "Infiltrating Tumor (IT)",
    "normalized_area_ithbv" =
    "Hyperplastic blood vessels in infiltrating tumor (IThbv)",
    "normalized_area_ct"   =   
    "Cellular Tumor (CT)",
    "normalized_area_ctpnz" =
    "Perinecrotic zone (CTpnz)",
    "normalized_area_cthbv" =
    "Hyperplastic blood vessels in cellular tumor (CThbv)",
    "normalized_area_ctpnn" =
    "Pseudopalisading cells but no visible necrosis (CTpnn)",
    "normalized_area_ctpan" =
    "Pseudopalisading cells around necrosis (CTpan)",
    "normalized_area_ctmvp" =
    "Microvascular proliferation (CTmvp)",
    "normalized_area_ctne" =
    "Necrosis (CTne)")
    
    molmap = structure(c("NA", "C", "CM", "CN", "M", "MN", "N", "NP", "P"), .Names = c("", 
       "Classical", "Classical, Mesenchymal", "Classical, Neural", "Mesenchymal", 
       "Mesenchymal, Neural", "Neural", "Neural, Proneural", "Proneural"
       ))
    
    server = function(input, output, session) {
     sb = metadata(ivySE)$subBlockDetails
     sb = sb[!is.na(sb$survival_days),]
     output$xyplot = renderPlotly({
       df = sb[, c(input$x, input$y, "donor_id", "molecular_subtype",
                    "specimen_page_link")]
       df$donor_id = paste0("donor: ", df$donor_id)
       df$molm = molmap[df$molecular_subtype]
       p = ggplot(df, aes_(x=as.name(input$x), y=as.name(input$y), text=as.name("donor_id"))) + 
              geom_point() #data=df, mapping=aes_(colour=as.name("molm")))
       gp = ggplotly(p, source="subset", tooltip="text") %>% layout(dragmode="select") #plot(sb[, input$x], sb[, input$y], xlab=input$x, ylab=input$y )
       event.data <- event_data("plotly_click", source = "subset")
       if (!is.null(event.data)) browseURL(df[event.data$pointNumber+1,
           "specimen_page_link"])
       gp
       })
     
    procSel = reactive({
        event.data <- event_data("plotly_selected", source = "subset")
        if(is.null(event.data) == TRUE) return(NULL)
        dr = duplicated(sb$donor_id)
        udf <<- sb[-which(dr),]  # survfit does not find without <<-
        udf$grp <<- 0
        sdf = sb[event.data$pointNumber+1,]
        indo = sdf$donor_id
        udf[which(udf$donor_id %in% indo),]$grp <<- 1
        survfit(Surv(survival_days, rep(1,nrow(udf)))~grp, data=udf)
        })
     
     output$plot2 = renderPlot({
        validate(need(!is.null(procSel()), "waiting for (dragged) selection"))
        mm = procSel() #survfit(Surv(survival_days, rep(1,nrow(udf)))~grp, data=udf)
        suppressWarnings({ ggsurvplot(mm) })
       })
    
     output$boxes1 = renderPlot({
        # Get subset based on selection
        event.data <- event_data("plotly_selected", source = "subset")
        # If NULL dont do anything
        if(is.null(event.data) == TRUE) return(NULL)
        dr = duplicated(sb$donor_id)
        udf <<- sb[-which(dr),]
        udf$grp = 0
        sdf = sb[event.data$pointNumber+1,]
        intu = sdf$tumor_name
        seSEL = ivySE[someSets[[input$gs]], which(ivySE$tumor_name %in% intu)]
        logp = function(x) log(x+1)
        par(mar=c(5,4,2,2))
        meds = apply(assay(seSEL),1,median,na.rm=TRUE)
        omeds = order(meds)
        boxplot( data.frame(logp(t(assay(seSEL)[omeds,]))), main="in image subset", las=2, ylab="log fpkm",
          ylim=c(0,7))
       })
    
     output$boxes2 = renderPlot({
        # Get subset based on selection
        event.data <- event_data("plotly_selected", source = "subset")
        # If NULL dont do anything
        if(is.null(event.data) == TRUE) return(NULL)
        dr = duplicated(sb$donor_id)
        udf <<- sb[-which(dr),]
        udf$grp = 0
        sdf = sb[event.data$pointNumber+1,]
        intu = sdf$tumor_name
        seSEL = ivySE[someSets[[input$gs]], which(ivySE$tumor_name %in% intu)] # for ordering
        meds = apply(assay(seSEL),1,median,na.rm=TRUE)
        omeds = order(meds)
        seUNSEL = ivySE[someSets[[input$gs]], -which(ivySE$tumor_name %in% intu)]
        logp = function(x) log(x+1)
        par(mar=c(5,4,2,2))
        boxplot( data.frame(logp(t(assay(seUNSEL)[omeds,]))), main="not in image subset", las=2, ylab="log fpkm",
            ylim=c(0,7))
       })
      output$vocab = renderTable({
        plop = function(x) gsub("^", " ", x)
        data.frame(short=c("_Molec. subtype_", names(molmap), "_Feature_", names(featanno)), 
                 long=c(" ", as.character(molmap), " ", as.character(featanno)))
        })
      output$bkgrd = renderText( paste(metadata(ivySE)$README, collapse="\n") )
      output$versinfo = renderText( paste0("ivygapSE version ", 
                              as.character(packageVersion("ivygapSE"))) )
      output$commentary = renderText( paste(
           "This app introduces users to quantitative elements of the Ivy Glioblastoma Atlas project.",
           "Tumors are divided into blocks and subblocks.  Subblocks are associated with",
           "image-derived measures such as proportions of cells deemed infiltrating or cellular,",
           "and with measures of gene expression derived from RNA-seq on 270=122+148 samples",
           "microdissected after selection anatomic and molecular criteria.",
           "'Donor' refers to the individual patient, and survival times are recorded for most patients.",
           "This app allows interactive selection of (a) image features for scatterplotting,",
           "(b) image sets for stratified survival distribution estimation,",
           "and (c) gene sets for expression distribution comparison between strata.",
           "More work is needed to define an interface that distinguishes origins of expression assays.",
           "Data incompleteness is substantial and no effort is made to support",
           "statistical hypothesis testing with this app.",
           "The following upset diagram (see http://caleydo.org/tools/upset/) illustrates",
           "data availability configurations for the image-derived tumor features.") )
      output$upset = renderPlot({
         md = metadata(ivySE)$subBlock
         mm = md[,16:26]
         upset(data.frame(1-is.na(mm)), 11)
         })
     }
    shinyApp(ui=ui, server=server)
}