FlowJo workspace import

[schienstockd]

Hello,

You have a cool framework - I was just wondering whether I could import a FlowJo Workspace directly into spectre?

I have seen your tutorial:
https://wiki.centenary.org.au/display/SPECTRE/Exporting+data+from+FlowJo+for+analysis+in+Spectre

Could we utilise flowWorkspace somehow?
https://github.com/RGLab/flowWorkspace

Thanks!

Dominik

[tomashhurst]

Hi @schienstockd , sorry for the late reply! I have actually been thinking about this quite a bit. I am keen to get something like this working, and I'm sure it can be done, but we are a bit tied up with some other major developments at the moment. Are you using flowWorkspace at the moment?

[schienstockd]

Yes, I analyse my samples in R with flowWorkspace, openCyto and ggcyto. Most people in the lab use FlowJo for gating but some would like to use R for further analysis. It would be good to make this transition relatively painless.

How can I load data into spectre? Do I need to save the populations as CSV or could I also push a list of data.tables somehow?
Something like this?

library(CytoML)
library(ggcyto)

# read FlowJo workspace
wsfile <- "path/to/workspace.wsp"
ws <- open_flowjo_xml(wsfile)

# get second group
gs <- flowjo_to_gatingset(ws, name = 2)

# get a list of data.tables
subsets <- c("T cells", "B cells", "DC")

wsSubsets <- list()
for (subset in subsets) {
  # fortify gating set
  attr(gs, "subset") <- subset
  wsSubsets[[subset]] <- fortify(gs)
}

# How to open that in spectre.. ?

[tomashhurst]

It should very much be possible to extract a list of data.tables, as you have indicated above. Spectre essentially just manages data as a large table of cells (rows) vs features/markers (columns). So you can pull the matrix of values for each file, and merge them together into a data.table and just proceed straight into any of the Spectre workflows. This would be a great addition, as it would be handy way of rapidly pulling in pre-labelled data via gating (rather than going to the effort of exporting the gates etc).

I'll try out this approach on Monday, as you've peaked my curiosity, but you might get an answer before that. What exactly does the fortify function gives you in your loop?

[schienstockd]

fortify is a function from the ggcyto package. They implemented it primarily for plotting the data from a GatingSet with ggplot, I think.

It will create a data.table with the channel intensities from the FCS file and the associated pData.
I am not sure whether extra information from the FlowJo workspace will also be included in there, for example, when you assign specific treatments or timepoints to the individual tubes/wells.

[denvercal1234GitHub]

Would you mind helping me understand the advantage of importing the whole workspace versus what is being described here (https://wiki.centenary.org.au/display/SPECTRE/Exporting+data+from+FlowJo+for+analysis+in+Spectre)? I thought exporting gated population of interest from each sample as csv would give me more control how to process these data before downstream analysis? Thank you all!

[schienstockd]

The idea is to automate the export of gated populations from the workspace - not to import the whole workspace. The advantage would be that if you change the gating in your workspace you would only have to run this import script/function and have the changed populations in spectre without having to manually export them again from FlowJo.

It seems that you could pull them from the workspace, create a list of data.tables, then merge them into one big data.table and use the populations in spectre.

[SamGG]

The idea is very interesting. Making the process from workspace to spectre as direct as possible would attract a lot of analysts. And FlowJo is used a lot.

[tomashhurst]

It would potentially be advantageous when comparing gates and and clusters, and being able to select gated subsets in a flexible way during analysis is quite appealing -- rather than exporting specific gates etc and then going back and forth through that process.

[schienstockd]

Here is a naive start. Then you will have all populations in one data.table.

library(CytoML)
library(ggcyto)
library(data.table)

# Load FlowJo workspace
load.flowjo.gs <- function(wsp.file, sample.group = 1) {
  # read FlowJo workspace
  ws <- open_flowjo_xml(wsp.file)
  
  # get sample group
  gs <- flowjo_to_gatingset(ws, name = sample.group)
  
  gs
}

# get gated populations from FlowJo workspace
gated.pops.to.data.table <- function(gs, gated.pops) {
  # get subsets
  subset.list <- lapply(gated.pops, function(x) {
    # fortify gating set
    attr(gs, "subset") <- x
    y <- fortify(gs)
    
    # add population name
    y[, ("FlowPop") := x]
    
    y
  })
  
  # bind 
  DT <- rbindlist(subset.list)
  
  # prepare data table
  DT[, c(".rownames") := NULL]
  setnames(DT, "name", "FileName")
  DT[, FileNo := .GRP, by = .(FileName)]
  
  DT
}

# load workspace
flowjo.gs <- load.flowjo.gs("path/to/workspace.wsp", sample.group = 2)

# get population table
gated.pops <- gated.pops.to.data.table(flowjo.gs, c("cDC1", "cDC2", "pDC"))

[tomashhurst]

Badass! I'm definitely giving this a shot later

[schienstockd]

You might have to checkout the reference for flowjo_to_gatingset. There are a few options to specify the FCS path, transformation and compensation.