scale_x_logicle() not converging anymore

[phauchamps]

Hi Mike and team,

I am currently in the process of submitting a new package to Bioconductor, CytoPipeline, which depends on ggcyto::scale_x_logicle().

Since the new Bioc 3.16 release, and in 3.17 devel, this function does not work anymore, regardless (it seems) of the inputs I provide to it. I constantly receive error messages like the following:

Error: Logicle Exception: T = 262144.000000, W = 2.000000, M = 6.420000, A = 0.000000
   DidNotConverge: scale() didn't converge

I can see that the same error occurs in ggcyto vignette building in the Bioc builds (see e.g. here )

Have you already identified where this problem come from ? Any plan to have this solved soon ?

Thanks a lot,

Philippe

[gfinak]

We'll investigate, in the meantime have a look here for a possible solution
RGLab/flowCore#41

[phauchamps]

Hi Greg,
Yes I knew already about the post you are pointing to.
However, what we have here is a rather serious issue with the logicle function from flowCore, since even default parameters are generating this convergence error, and the ggcyto vignette does not work anymore.

scale_x_flowJo_biexp(), on the other hand, while not generating errors, is no option for me, as it does not look the same, and does not correspond to the parameters obtained using flowCore::estimateLogicle() - unless you know of an exact one to one correspondance between the flowjo_biexp() parameters and the logicle() parameters: I tried myself, for the flowjo_biexp parameters :

maxValue = logicleParams$t,
widthBasis = -10^logicleParams$w,
pos = logicleParams$m - logicleParams$w,
neg = logicleParams$a

[mikejiang]

can't seem to reproduce it locally, my bioconductor docker image is 5-week old, let me pull the latest to try again

[mikejiang]

Now I am on the latest bioc devel, can reproduce the same error

> library(ggcyto)
> data(GvHD)
> fr <- GvHD[[1]]
> p <- autoplot(fr, "FL1-H")
> p + scale_x_logicle() #flowCore logicle scale
Error: Logicle Exception: T = 262144.000000, W = 0.500000, M = 4.500000, A = 0.000000
DidNotConverge: scale() didn't converge

However, applying the same transformation to the same data outside of ggcyto, it succeeds

> lgcl <- logicleTransform()
> trans <- transformList(c("FL1-H"), lgcl)
> after <- transform(fr, trans)
> range(fr)["FL1-H"]
    FL1-H
min     1
max 10000
> range(after)["FL1-H"]
        FL1-H
min 0.5050419
max 3.0772470

[phauchamps]

Thank you Mike, please let me know asap when you have found the cause of this error :-)

[flbgmazo]

I am having the same problem. Just stated appearing today. It was working perfectly in the morning but the issue appeared in the afternoon after I installed other packages (ggpubr, several dependencies were also updated/installed). Since, scale_x_logicle() is not critical, I tried excluding it. Although the error does not occur if scale_x_logicle() is removed, the resulting plots are empty with no data. Help!!!

[mikejiang]

It seems to be introduced by the latest ggplot 3.4.0 release, not sure if it is a bug or breaking change from it. Still investigating...
In the meantime, you can downgrade ggplot to 3.3.x which should resolve your issue for now

[flbgmazo]

I downgraded and it worked (at least for the first few plots that I tried). Have a great weekend.

[flbgmazo]

Thanks for working on this amazing package

[mikejiang]

@phauchamps flowjo_biexp should be equivalent to logicle and is more robust.
The correct mapping between two is

widthBasis = -10^(2 * w)
maxRange=t
neg=a
pos=m

In the latest patch, I just replaced logicle_trans with flowjo_biexp_trans in scale_x_logicle() function so that it is free from DidNotConverge exception.
Here is the plot now using biexp_trans under the hood

[phauchamps]

Thank you Mike. However, are you sure of the parameters conversion from logicle to flojo_biexp(). I tried using the new github install of ggcyto, or by calling myself the scale_x_flowjo_biexp in my code using the same parameter conversion, and I get very different results than before. Actually, the new versions of the plots do not look very good.

[mikejiang]

without the example code and data, we can tell where went wrong

[phauchamps]

Hi Mike, I downgraded to ggplots2 3.3.6 in order to retrieve the previous scale_x_logicle() results.
Here is a minimal reproduceable example, using fcs file in .zip in attachment

library(ggcyto)

# TO UPDATE
datasetPath <- "/Datasets"
fileName <- "sample_Donor1.fcs"

mySample <- read.FCS(filename = paste0(datasetPath, "/", fileName),
                     transformation = "linearize",
                     alter.names = FALSE,
                     name = "OMIP-21",
                     truncate_max_range = TRUE,
                     min.limit = NULL)



a=1
w=2
[IssueLogicle_20221205.zip](https://github.com/RGLab/ggcyto/files/10152759/IssueLogicle_20221205.zip)
[IssueLogicle_20221205.zip](https://github.com/RGLab/ggcyto/files/10152763/IssueLogicle_20221205.zip)

m=7
t=262144

xChannel <- "450/50Violet-A"

p <- ggplot(
    data = mySample,
    mapping = aes(x = .data[[xChannel]])
) +
    geom_density(fill = "lightblue", alpha = 0.2, na.rm = TRUE) +
    xlab(xChannel)

p + scale_x_logicle(a = a,
                    w = w,
                    m = m,
                    t = t) +
    ggtitle("scale_x_logicle")
                         

p + scale_x_flowjo_biexp(maxValue = t,
                         widthBasis = -10^(2*w),
                         pos = m,
                         neg = a) +
    ggtitle("scale_x_flowjo_biexp")
                                                     
   

When using ggcyto 1.26.0 (Bioc version), I get the first plot with scale_x_logicle(), and the second plot with scale_x_flowjo_biexp(). When using last ggcyto version from repo, I get twice the second plot.

I think the parameters matching is not good.