callModule(modulePopover,"modulePopover_convertChooseDatafile",
data = reactive(list(title = "Data file",
content="Select one (.txt, .csv, .tsv, .xls, .xlsx) file.")))
callModule(modulePopover,"modulePopover_convertIdType",
data = reactive(list(title = "ID definition",
content="If you choose the automatic ID, Prostar will build an index.")))
callModule(modulePopover,"modulePopover_convertProteinID",
data = reactive(list(title = "Select protein IDs",
content="Select the column containing the parent protein IDs.")))
callModule(modulePopover,"modulePopover_convertDataQuanti",
data = reactive(list(title = "Quantitative data",
content="Select the columns that are quantitation values by clicking in the field below.")))
mod_staticDT_server("overview_convertData",
data = reactive({GetDatasetOverview()})
)
##--------------------------------------------------------------
## Gestion du slideshow
##--------------------------------------------------------------
output$checkConvertPanel <- renderUI({
rv$tab1
rv$pageConvert
color <- rep("lightgrey",NUM_PAGES_CONVERT)
txt <- c("Select file", "Select ID", "Select quantitative data", "Build design", "Convert")
buildTable(txt, color)
})
########### STEP 1 ############
output$Convert_SelectFile <- renderUI({
tagList(br(), br(),
radioButtons("choose_software", "Software to import from",
choices = setNames(nm = c('maxquant', 'proline')),
selected = character(0)
),
uiOutput('choose_file_to_import'),
uiOutput("ManageXlsFiles"),
uiOutput("ConvertOptions")
)
})
output$choose_file_to_import <- renderUI({
req(input$choose_software)
fluidRow(
column(width=2, modulePopoverUI("modulePopover_convertChooseDatafile")),
column(width = 10, fileInput("file1", "",
multiple=FALSE,
accept=c(".txt", ".tsv", ".csv",".xls", ".xlsx"))))
})
fileExt.ok <- reactive({
req(input$file1$name)
authorizedExts <- c("txt","csv", "tsv","xls","xlsx")
ext <- GetExtension(input$file1$name)
!is.na(match(ext, authorizedExts))
})
output$ConvertOptions <- renderUI({
req(input$choose_software)
req(input$file1)
req(fileExt.ok())
tagList(
radioButtons("typeOfData",
"Is it a peptide or protein dataset ?",
choices=c("peptide dataset" = "peptide",
"protein dataset" = "protein")
)
,radioButtons("checkDataLogged",
"Are your data already log-transformed ?",
#width = widthWellPanel,
choices=c("yes (they stay unchanged)" = "yes",
"no (they wil be automatically transformed)"="no"),
selected="no")
,br()
,checkboxInput("replaceAllZeros",
"Replace all 0 and NaN by NA",
value= TRUE)
)
})
############ Read text file to be imported ######################
observeEvent(c(input$file1, input$XLSsheets),{
input$XLSsheets
if (((GetExtension(input$file1$name) %in% c("xls","xlsx")))
&& is.null(input$XLSsheets)) {return(NULL) }
authorizedExts <- c("txt","csv", "tsv","xls","xlsx")
if (!fileExt.ok()){
shinyjs::info("Warning : this file is not a text nor an Excel file !
Please choose another one.")
} else {
# result = tryCatch(
# {
ClearUI()
ClearMemory()
ext <- GetExtension(input$file1$name)
shinyjs::disable("file1")
switch(ext,
txt = { rv$tab1 <- read.csv(input$file1$datapath, header=TRUE, sep="\t", as.is=T)},
csv = { rv$tab1 <- read.csv(input$file1$datapath, header=TRUE, sep=";", as.is=T)},
tsv = { rv$tab1 <- read.csv(input$file1$datapath, header=TRUE, sep="\t", as.is=T)},
xls = { rv$tab1 <- readExcel(input$file1$datapath, ext, sheet=input$XLSsheets)},
xlsx = {rv$tab1 <- readExcel(input$file1$datapath, ext, sheet=input$XLSsheets)}
)
colnames(rv$tab1) <- gsub(".", "_", colnames(rv$tab1), fixed=TRUE)
colnames(rv$tab1) <- gsub(" ", "_", colnames(rv$tab1), fixed=TRUE)
# }
# , warning = function(w) {
# shinyjs::info(conditionMessage(w))
# }, error = function(e) {
# shinyjs::info(paste("Read text file to convert",":",
# conditionMessage(e),
# sep=" "))
# }, finally = {
# #cleanup-code
# })
}
#shinyjs::disable('file1')
})
output$ManageXlsFiles <- renderUI({
req(input$choose_software)
req(input$file1)
.ext <- GetExtension(input$file1$name)
if ((.ext == "xls") || (.ext == "xlsx")){
sheets <- listSheets(input$file1$datapath)
selectInput("XLSsheets", "sheets", choices = as.list(sheets),
width='200px')
}
})
################## STEP 2 ###############################
output$Convert_DataId <- renderUI({
tagList(
br(), br(),
tags$div(
tags$div( style="display:inline-block; vertical-align: top; padding-right: 100px;",
uiOutput("id"),
uiOutput("warningNonUniqueID")
),
tags$div( style="display:inline-block; vertical-align: top;",
uiOutput("convertChooseProteinID_UI"),
uiOutput("previewProteinID_UI")
)
)
)
})
output$id <- renderUI({
req(rv$tab1)
.choices <- c("AutoID", colnames(rv$tab1))
names(.choices) <- c("Auto ID", colnames(rv$tab1))
tagList(
modulePopoverUI("modulePopover_convertIdType"),
selectInput("colnameForID", label = "", choices = .choices)
)
})
output$warningNonUniqueID <- renderUI({
req(input$colnameForID != 'AutoID')
req(rv$tab1)
t <- (length(as.data.frame(rv$tab1)[, input$colnameForID])
== length(unique(as.data.frame(rv$tab1)[, input$colnameForID])))
if (!t){
text <- "<img src=\"images/Problem.png\" height=\"24\"></img><font color=\"red\">
Warning ! Your ID contains duplicate data.
Please choose another one."
}
else {
text <- "<img src=\"images/Ok.png\" height=\"24\"></img>"
}
HTML(text)
})
output$convertChooseProteinID_UI <- renderUI({
req(rv$tab1)
req(input$typeOfData != "protein")
.choices <- c("",colnames(rv$tab1))
names(.choices) <- c("",colnames(rv$tab1))
tagList(
modulePopoverUI("modulePopover_convertProteinID"),
selectInput("convert_proteinId",
"",
choices = .choices , selected = character(0))
)
})
output$helpTextDataID <- renderUI({
input$typeOfData
if (is.null(input$typeOfData)){return(NULL)}
t <- ""
switch(input$typeOfData,
protein = {t <- "proteins"},
peptide = {t <- "peptides"}
)
txt <- paste ("Please select among the columns of your data the one that
corresponds to a unique ID of the ", t, ".", sep=" ")
helpText(txt)
})
datasetID_Ok <- reactive({
req(input$colnameForID)
req(rv$tab1)
if (input$colnameForID == "AutoID") {t <- TRUE}
else {
t <- (length(as.data.frame(rv$tab1)[, input$colnameForID])
== length(unique(as.data.frame(rv$tab1)[, input$colnameForID])))
}
t
})
output$previewProteinID_UI <- renderUI({
req(input$convert_proteinId != "")
tagList(
p(style="color: black;", 'Preview'),
tableOutput("previewProtID")
)
})
output$previewProtID <- renderTable(
# req(input$convert_proteinId),
head(rv$tab1[,input$convert_proteinId]),
colnames = FALSE
)
##################### STEP 3 #######################################
# The code for MaxQuant import is in the file srv_ConvertData_MaxQuant.R
# The code for Proline import is in the file srv_ConvertData_Proline.R
# output$select_quanti_data <- renderUI({
# fluidRow(
# column(width=4, uiOutput("eData",width = "400px")),
# column(width=8, shinyjs::hidden(
# DT::dataTableOutput("x1", width='500px'))
# )
# )
# })
#
# output$eData <- renderUI({
# input$file1
# req(rv$tab1)
#
# choices <- colnames(rv$tab1)
# names(choices) <- colnames(rv$tab1)
#
# tagList(
# modulePopoverUI("modulePopover_convertDataQuanti"),
# selectInput("choose_quantitative_columns",
# label = "",
# choices = choices,
# multiple = TRUE, width='200px',
# size = 20,
# selectize = FALSE)
# )
# })
#
#
# output$Convert_ExpFeatData <- renderUI({
#
# tagList(
# shinyjs::useShinyjs(),
# fluidRow(
# column(width=4, shinyjs::hidden(
# checkboxInput("selectIdent",
# "Select columns for identification method",
# value = FALSE))
# )
# ),
# uiOutput('select_quanti_data'),
# tags$script(HTML("Shiny.addCustomMessageHandler('unbind-DT', function(id) {
# Shiny.unbindAll($('#'+id).find('table').DataTable().table().node());
# })"))
# )
# })
#--------------------------
output$Convert_ExpFeatData <- renderUI({
tagList(
shinyjs::useShinyjs(),
fluidRow(
column(width = 4, radioButtons("selectIdent","Provide identification method",
choices = list("No (default values will be computed)" = FALSE,
"Yes" = TRUE),
selected = FALSE
)
),
column(width = 4,uiOutput("checkIdentificationTab")),
column(width = 4, shinyjs::hidden(
div(id = 'warning_neg_values',
p("Warning : Your original dataset may contain negative values",
"so that they cannot be logged. Please check back the dataset or",
"the log option in the first tab."))
)
)
),
fluidRow(
column(width = 4, uiOutput("eData", width = "400px")),
column(width = 8, shinyjs::hidden(
uiOutput('inputGroup', width='600px'))
)
)
#)
# tags$script(HTML("Shiny.addCustomMessageHandler('unbind-DT', function(id) {
# Shiny.unbindAll($('#'+id).find('table').DataTable().table().node());
# })"))
)
})
output$inputGroup = renderUI({
#if (is.null(input$choose_quantitative_columns) || is.null(rv$tab1))
# return(NULL)
n <- length(input$choose_quantitative_columns)
input_list <- lapply(seq_len(n), function(i) {
inputName <- paste("colForOriginValue_", i, sep = "")
div(
div( style="align: center;display:inline-block; vertical-align: middle;padding-right: 10px;",
p(tags$strong(paste0('Identification col. for ', input$choose_quantitative_columns[i])))
),
div( style="align: center;display:inline-block; vertical-align: middle;",
selectInput(inputName, '', choices = c('None', colnames(rv$tab1)))
)
)
})
do.call(tagList, input_list)
})
observeEvent(input[['colForOriginValue_1']], ignoreInit = T, ignoreNULL = F, {
n <- length(input$choose_quantitative_columns)
lapply(seq(2,n), function(i) {
inputName <- paste("colForOriginValue_", i, sep = "")
start <- which(colnames(rv$tab1)==input[['colForOriginValue_1']])
if (input[['colForOriginValue_1']] == 'None')
.select <- 'None'
else
.select <- colnames(rv$tab1)[(i-1)+start]
updateSelectInput(session, inputName, selected = .select)
})
})
observe({
shinyjs::toggle('warning_neg_values', condition = !is.null(input$choose_quantitative_columns) && length(which(rv$tab1[,input$choose_quantitative_columns] < 0)) > 0)
shinyjs::toggle('selectIdent', condition = !is.null(input$choose_quantitative_columns))
shinyjs::toggle('inputGroup', condition = as.logical(input$selectIdent) == TRUE)
})
output$eData <- renderUI({
input$file1
req(rv$tab1)
choices <- colnames(rv$tab1)
names(choices) <- colnames(rv$tab1)
tagList(
modulePopoverUI("modulePopover_convertDataQuanti"),
selectInput("choose_quantitative_columns",
label = "",
choices = choices,
multiple = TRUE, width='200px',
size = 20,
selectize = FALSE)
)
})
output$checkIdentificationTab <- renderUI({
req(as.logical(input$selectIdent) == TRUE)
shinyValue("colForOriginValue_",length(input$choose_quantitative_columns))
temp <- shinyValue("colForOriginValue_",length(input$choose_quantitative_columns))
# if ((length(which(temp == "None")) == length(temp)))
# {
# img <- "images/Ok.png"
# txt <- "Correct"
# } else {
if (length(which(temp == "None")) > 0)
{
img <- "images/Problem.png"
txt <- "The identification method is not appropriately defined for each sample."
} else {
if(length(temp) != length(unique(temp))){
img <- "images/Problem.png"
txt <- "There are duplicates in identification columns."
}else {
img <- "images/Ok.png"
txt <- "Correct"
}
}
#}
tags$div(
tags$div(
tags$div(style="display:inline-block;",tags$img(src = img, height=25)),
tags$div(style="display:inline-block;",tags$p(txt))
)
)
})
# GetToto <- reactive({
# if (is.null(input$choose_quantitative_columns) || is.null(rv$tab1)) return(NULL)
# shinyValue("colForOriginValue_",1)
# isolate({
# ind <- which(shinyValue("colForOriginValue_", 1) == colnames(rv$tab1))
# n <- nrow(as.data.frame(input$choose_quantitative_columns))
# if (length(ind) > 0)
# colnames(rv$tab1)[ind:(ind+n)]
# else
# rep('', n)
# })
# })
# reactive dataset
# quantiDataTable <- reactive({
# # req(c(input$eData.box,rv$tab1))
# # as.logical(input$selectIdent)
#
# if (is.null(input$choose_quantitative_columns) || is.null(rv$tab1)) return(NULL)
#
# df <- NULL
# session$sendCustomMessage('unbind-DT', 'x1')
# choices <- c("None", colnames(rv$tab1))
# names(choices) <- c("None", colnames(rv$tab1))
#
#
# if (isTRUE(as.logical(input$selectIdent))) {
# browser()
# # df <- data.frame(as.data.frame(input$choose_quantitative_columns),
# # shinyInput(selectInput,
# # "colForOriginValue_",
# # nrow(as.data.frame(input$choose_quantitative_columns)),
# # choices = choices)
# # )
#
# df <- data.frame(as.data.frame(input$choose_quantitative_columns),
# shinyInput(selectInput,
# id = "colForOriginValue_1",
# num = 1,
# choices = choices,
# selected = colnames(rv$tab1)[30]),
# shinyInput(selectInput,
# "colForOriginValue_2",
# num = 1,
# choices = choices,
# selected = eventReactive(shinyValue("colForOriginValue_1",1), {
# return(colnames(rv$tab1)[33])
# })),
# shinyInput(selectInput,
# "colForOriginValue_3",
# num = 1,
# choices = choices,
# selected = colnames(rv$tab1)[32]),
# shinyInput(selectInput,"colForOriginValue_4",
# num = 1,
# choices = choices,
# selected = colnames(rv$tab1)[33]),
# shinyInput(selectInput,
# "colForOriginValue_5",
# num = 1,
# choices = choices,
# selected = colnames(rv$tab1)[34]),
# shinyInput(selectInput,"colForOriginValue_6",
# num = 1,
# choices = choices,
# selected = colnames(rv$tab1)[35]),
# shinyInput(selectInput,"colForOriginValue_7",
# num = 1,
# choices = choices,
# selected = colnames(rv$tab1)[36]),
# shinyInput(selectInput,"colForOriginValue_8",
# num = 1,
# choices = choices,
# selected = colnames(rv$tab1)[37])
# )
#
#
#
# colnames(df) <- c("Sample", "Identification method")
# } else {
# df <- data.frame(Sample = as.data.frame(input$choose_quantitative_columns))
# colnames(df) <- c("Sample")
# }
# df
# })
# output$x1 <- renderDataTable(
# quantiDataTable(),
# escape=FALSE,
# rownames = FALSE,
# extensions = c('Scroller'),
# server = FALSE,
# selection = 'none',
# class = 'compact',
# options = list(
# preDrawCallback=JS(
# 'function() {
# Shiny.unbindAll(this.api().table().node());}'),
# drawCallback= JS(
# 'function(settings) {
# Shiny.bindAll(this.api().table().node());}'),
# # rowCallback = JS("function(r,d) {$(r).attr('height', '10px')}"),
# dom = 'frtip',
# autoWidth = TRUE,
# deferRender = TRUE,
# bLengthChange = FALSE,
# scrollX = 200,
# scrollY = 500,
# scroller = TRUE,
# ajax = list(url = dataTableAjax(session, quantiDataTable()))
#
# )
#
# )
#
#
# observeEvent(shinyValue("colForOriginValue_",
# nrow(as.data.frame(quantiDataTable()))),{})
checkIdentificationMethod_Ok <- reactive({
res <- TRUE
tmp <- NULL
if (isTRUE(as.logical(input$selectIdent))) {
tmp <- shinyValue("colForOriginValue_",nrow(quantiDataTable()))
if ((length(grep("None", tmp)) > 0) || (sum(is.na(tmp)) > 0)){ res <- FALSE }
}
res
})
############# STEP 4 ######################
output$Convert_BuildDesign <- renderUI({
req(input$file1)
tagList(
tags$p("If you do not know how to fill the experimental design, you can click
on the '?' next to each design in the list that appear once the conditions
are checked or got to the ",
actionLink("linkToFaq1", "FAQ",style="background-color: white"),
" page."),
fluidRow(
column(width=6,tags$b("1 - Fill the \"Condition\" column to identify the conditions to compare.")),
column(width=6,uiOutput("UI_checkConditions") )
),
fluidRow(
column(width=6,uiOutput("UI_hierarchicalExp")),
column(width=6,uiOutput("checkDesign") )
),
hr(),
selectInput("convert_reorder", "Order by conditions ?",
choices=c("No"="No", "Yes"="Yes"),
width="100px"),
tags$div(
tags$div(style="display:inline-block; vertical-align: top;",
uiOutput("viewDesign",width="100%")
),
tags$div(style="display:inline-block; vertical-align: top;",
shinyjs::hidden(div(id = "showExamples", uiOutput("designExamples") ))
)
)
)
})
############# STEP 5 ########################
output$Convert_Convert <- renderUI({
tagList(
br(), br(),
uiOutput("convertFinalStep"),
mod_staticDT_ui("overview_convertData"),
uiOutput("conversionDone"),
p("Once the 'Load' button (above) clicked, you will be automatically redirected to Prostar home page. The dataset will be accessible within Prostar
interface and processing menus will be enabled. However, all importing functions ('Open MSnset', 'Demo data' and 'Convert data') will be disabled
(because successive dataset loading can make Prostar unstable). To work on another dataset, use first the 'Reload Prostar' functionality from
the 'Dataset manager' menu: it will make Prostar restart with a fresh R session where import functions are enabled.")
)
})
output$convertFinalStep <- renderUI({
req(rv$designChecked)
if (!(rv$designChecked$valid)){return(NULL)}
tagList(
uiOutput("checkAll_convert", width="50"),
htmlOutput("msgAlertCreateMSnset"),
hr(),
textInput("filenameToCreate","Enter the name of the study"),
actionButton("createMSnsetButton","Convert data", class = actionBtnClass),
uiOutput("warningCreateMSnset")
)
})
output$conversionDone <- renderUI({
rv$current.obj
if (is.null(rv$current.obj)) { return(NULL)}
h4("The conversion is done. Your dataset has been automatically loaded
in memory. Now, you can switch to the Descriptive statistics panel to
vizualize your data.")
})
# updateInputs <- function(id, n){
# for (i in seq_len(n)) {
# updateSelectInput(paste0(id,i),label=NULL,selected = input[[paste0(id,i)]])
# }
# }
output$warningCreateMSnset <- renderUI({
if (isTRUE(as.logical(input$selectIdent))){
n <- length(input$choose_quantitative_columns)
colNamesForMetacell <- unlist(lapply(seq_len(n), function(x) {
input[[paste0('colForOriginValue_', x)]]
}))
if (length(which(colNamesForMetacell == "None")) >0){
text <- "<font color=\"red\"> Warning: The MSnset cannot be created because the identification
method are not fully filled. <br>"
HTML(text)
}
}
})
# observeEvent(input$fData.box,ignoreInit = TRUE,{
#
# choices = colnames(rv$tab1)[-which(colnames(rv$tab1) %in% input$fData.box)]
# names(choices) =
# colnames(rv$tab1)[-which(colnames(rv$tab1) %in% input$fData.box)]
# updateSelectInput(session, "choose_quantitative_columns",
# label = "",
# choices = choices,
# selected = choices)
#
# })
#######################################
observeEvent(input$createMSnsetButton,ignoreInit = TRUE,{
# if(is.null(input$createMSnsetButton) || (input$createMSnsetButton == 0))
#{return(NULL)}
#browser()
colNamesForMetacell <- NULL
if (isTRUE(as.logical(input$selectIdent))) {
n <- length(input$choose_quantitative_columns)
colNamesForMetacell <- unlist(lapply(seq_len(n), function(x) {
input[[paste0('colForOriginValue_', x)]]
}))
if (length(which(colNamesForMetacell == "None")) > 0 )
return (NULL)
if (!is.null(rv$newOrder))
colNamesForMetacell <- colNamesForMetacell[rv$newOrder]
}
isolate({
result = tryCatch(
{
ext <- GetExtension(input$file1$name)
txtTab <- paste("tab1 <- read.csv(\"", input$file1$name,
"\",header=TRUE, sep=\"\t\", as.is=T)", sep="")
txtXls <- paste("tab1 <- read.xlsx(",input$file1$name,
",sheet=", input$XLSsheets,")",sep="")
switch(ext,
txt = writeToCommandLogFile(txtTab),
csv = writeToCommandLogFile(txtTab),
tsv = writeToCommandLogFile(txtTab),
xls= writeToCommandLogFile(txtXls),
xlsx = writeToCommandLogFile(txtXls)
)
input$filenameToCreate
rv$tab1
tmp.choose_quantitative_columns <- input$choose_quantitative_columns
indexForEData <- match(tmp.choose_quantitative_columns, colnames(rv$tab1))
if (!is.null(rv$newOrder)){
tmp.choose_quantitative_columns <- tmp.choose_quantitative_columns[rv$newOrder]
indexForEData <- indexForEData[rv$newOrder]
}
indexForFData <- seq(1, ncol(rv$tab1))[-indexForEData]
# indexForIDBox <- NULL
# if (input$colnameForID != "AutoID") {
# indexForIDBox <- match(input$idBox, colnames(rv$tab1))
# }
#
metadata <- hot_to_r(input$hot)
logData <- (input$checkDataLogged == "no")
indexForMetacell <- NULL
if (!is.null(colNamesForMetacell) && (length(grep("None", colNamesForMetacell))==0) && (sum(is.na(colNamesForMetacell)) == 0)){
indexForMetacell <- match(colNamesForMetacell, colnames(rv$tab1))
}
options(digits=15)
protId <- NULL
if (input$typeOfData == 'protein')
protId <- input$colnameForID
else if(input$typeOfData == 'peptide')
protId <- input$convert_proteinId
tmp <- DAPAR::createMSnset(file = rv$tab1,
metadata = metadata,
indExpData = indexForEData,
colnameForID = input$colnameForID,
indexForMetacell = indexForMetacell,
logData = logData,
replaceZeros = input$replaceAllZeros,
pep_prot_data = input$typeOfData,
proteinId = gsub(".", "_", protId, fixed=TRUE),
software = input$choose_software
)
ClearUI()
ClearMemory()
rv$current.obj <- tmp
rv$current.obj.name <- input$filenameToCreate
rv$indexNA <- which(is.na(exprs(rv$current.obj)))
l.params <- list(filename = input$filenameToCreate)
loadObjectInMemoryFromConverter()
updateTabsetPanel(session, "tabImport", selected = "Convert")
},
# warning = function(w) {
# if (conditionMessage(w) %in% c("NaNs produced", "production de NaN")){
# shinyjs::info(paste("Warning : Your original dataset may contain negative values",
# "so that they cannot be logged. Please check back the dataset or",
# "the log option in the first tab.",
# sep=" "))
# } else {
# shinyjs::info(paste("Warning in CreateMSnSet",":",
# conditionMessage(w),
# sep=" "))
# }
# },
error = function(e) {
shinyjs::info(paste("Error :","CreateMSnSet",":",
conditionMessage(e),
sep=" "))
}, finally = {
#cleanup-code
})
})
})
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.