Use cellprofiler/cellprofiler:2.3.1 docker

[shntnu]

10

[shntnu]

I'm going to compare the outputs of 2.2.1 and 2.3.1 for one Cell Painting image, and if they are the same, will merge this. Does that sound reasonable @bethac07 ?

[bethac07]

👍

[shntnu]

Output from shntnu/cellprofiler:2.2.1 BR00100025-A01-1-v_2_2_1.zip

[shntnu]

Output from cellprofiler/cellprofiler:2.3.1 BR00100025-A01-1-v2_3_1.zip

[shntnu]

---

title: "Comparing CellProfiler v2.3.1 and v2.2.1" output: html_notebook

library(tidyverse)
library(magrittr)

Cells_221 <- read_csv("~/Desktop/BR00100025-A01-1-v_2_2_1/Cells.csv") %>% arrange(ImageNumber, ObjectNumber)
Cells_231 <- read_csv("~/Desktop/BR00100025-A01-1-v_2_3_1/Cells.csv") %>% arrange(ImageNumber, ObjectNumber)
Nuclei_221 <- read_csv("~/Desktop/BR00100025-A01-1-v_2_2_1/Nuclei.csv") %>% arrange(ImageNumber, ObjectNumber)
Nuclei_231 <- read_csv("~/Desktop/BR00100025-A01-1-v_2_3_1/Nuclei.csv") %>% arrange(ImageNumber, ObjectNumber)
Cytoplasm_221 <- read_csv("~/Desktop/BR00100025-A01-1-v_2_2_1/Cytoplasm.csv") %>% arrange(ImageNumber, ObjectNumber)
Cytoplasm_231 <- read_csv("~/Desktop/BR00100025-A01-1-v_2_3_1/Cytoplasm.csv") %>% arrange(ImageNumber, ObjectNumber)
Image_221 <- read_csv("~/Desktop/BR00100025-A01-1-v_2_2_1/Image.csv") 
Image_231 <- read_csv("~/Desktop/BR00100025-A01-1-v_2_3_1/Image.csv") 

Cells_diff <- compare::compare(Cells_221, Cells_231)

sort(names(Cells_diff$detailedResult[!Cells_diff$detailedResult]))

Cells_221 %<>% select(one_of(names(Cells_diff$detailedResult[!Cells_diff$detailedResult])))
Cells_231 %<>% select(one_of(names(Cells_diff$detailedResult[!Cells_diff$detailedResult])))

Cells_diff <- (Cells_221 - Cells_231)/Cells_221
Cells_diff %<>% 
  gather(variable, value)

p <- 
ggplot(Cells_diff,
       aes(abs(value))) + 
  geom_histogram() + 
  scale_x_log10() +
  facet_wrap(~variable) + 
  ggtitle("Comparing v2.3.1 and v2.2.1", 
          subtitle = "Distribution of relative absolute diff. between readouts of v2.3.1 and v2.2.1 across 481 cells. x-axis is on a log scale.")

ggsave("cells.png",  p,  width = 12, height = 10)

Nuclei_diff <- compare::compare(Nuclei_221, Nuclei_231)

sort(names(Nuclei_diff$detailedResult[!Nuclei_diff$detailedResult]))

Cytoplasm_diff <- compare::compare(Cytoplasm_221, Cytoplasm_231)

sort(names(Cytoplasm_diff$detailedResult[!Cytoplasm_diff$detailedResult]))

Image_diff <- compare::compare(Image_221, Image_231)

sort(names(Image_diff$detailedResult[!Image_diff$detailedResult]))

[shntnu]

[shntnu]

As shown above, there's a tiny difference between v2.2.1 and v2.3.1 – Zernike outputs are slightly different. No other features differ. The diff is likely due to https://github.com/CellProfiler/CellProfiler/pull/3455 If all this looks good to you, @bethac07 , I'll go ahead and merge.

[bethac07]

IIRC, Zernikes aren't that important, so I'm fine with it, but you being fine with it is much more important

[shntnu]

Thanks for reviewing. Turns out many Zernike's are pretty useful, but the 1% differences don't worry me. All set!

For reference: rug plot indicates reproducibility (measured as median rep correlation) of Zernikes. Histogram is the same measure but for all Cell Painting features.