Results not as predicated for cellular translocation


Hi CellProfiler team,

Thanks a lot for this open-sourcing program, which makes my imaging quantification possible.

I’m trying to process my imaging assay data acquired by automatic imaging plate reader by CellProfiler. As seen in my attached files, there’re two images stained by IF, one shown with protein enriched in plasma membrane and another with protein diffused in cytoplasm. And there’re two images with DAPI staining. I had a pipeline by which I shrunk cells for 4 pixels to identify plasma membrane and cytoplasm with the rest and then calculate the ratio between them. I feel segment of plasma membrane with this way is ok, however the results of ratio of PM to cytoplasm are only 1.14 and 1.02 respectively for those two image set. Then I tried to run your pipeline example with cytoplasm-nuclear location (ExampleVitra). For this example one image has staining in cytoplasma and another in nuclei, however the results as shown by ratio also gave a small difference. I also attach this example with image, pipeline and results. I did a little modification for this pipeline by adding illumCorrection in the same pipeline.

I’m confused with these results if I miss some points. I wonder if I should set a threshold to subtract background. Could you please review pipelines and have suggestions for me?

Thanks a lot.


My files locate in


Hi David,

In your CalculateMath, I see that you used MeanIntensity_PM / MeanIntensity_cytoplasm
(If I understand correctly, you meant “ShrunkenCell” is analogue for “cytoplasm” right?)

  • The “cytoplasm” defined in this pipeline is actually a flat 2-D cellular region that actually does contain a lot of membrane. Imagine a cell as a sphere, the surface of the sphere isn’t just its rim right?
    So the signals from membrane proteins are truly both high in your “PM” and “ShrunkenCell”. A ratio between them will be unsurprisingly small as you already noticed.
  • The above concern may be not easily solved, unless using a confocal microscope that can slice the focus plane thinner than thickness of the cell itself.
    But then, the results “1.14 and 1.02” would hint to us that, even with a limited resolution, there’s still a difference. And please note that you used “MeanIntensity” which is total signal normalized to area. So I would read those numbers as acceptable difference between protein accumulation in the rim-membrane (PM) v.s the flat “ShrunkenCell” region.

Hope that helps.


Hi Minh,

Thank you very much for your response. If I understand correctly, you meant the image also includes many pixels in Z-axis which is not focused in the image. This could be reasonable points. However if I calculate ratio between PM and cells (instead of shrunkencell, which is cells-PM), the difference is little as the same, 1.06 vs 1.01. The MeanIntensity of cells and PM is 0.159 and 0.168 in image 1(enriched in PM) and 0.109 and 0.110 in image 2 (not enriched in PM). Look like the difference of intensity in PM was not measured. Do you think if this is due to pixels that were not focused and seen in the image?

As the same, I’m also confused about the results of your example, named ExampleVitra, which I also uploaded in my folder. In this example channel 1-01 stained proteins in nuclei and channel 2-02 stained in cytoplasm. The IntegratedIntensity / MeanIntensity of cell, cytoplasm and nuclei are 60.52 / 0.172, 24.39 / 0.136 and 47.25 / 0.197 in image 01 and 69.16 / 0.191, 32.76 / 0.175 and 51.48 / 0.206 in image 02. Thus ratio of MeanIntensity of cytoplasm/cell and nuclei/cell are 0.80 and 1.14 in image 01 and 0.98 and 1.05 in image 02. To me the difference is so small, which is not as predicted. Could you please also look at what’s going on in this example?

Thanks a lot.




Finding a huge ratio difference between these is gonna be tricky, because your segmentation is never gonna be perfect based on how clumpy your cells are (but hopefully could be improved with some tinkering), you’re trying to hit a very fine region in the PM, you have the out-of-focus light as @Minh mentioned, and some of your cells (at least in the either seem not to have the Cy5 labeled protein at all or a whole clump will only have it at the exterior rather than at every edge). You can see what I mean in the image below (outlines are cell outlines, with Ratio1 values overlaid)- the cells with a pretty good segmentation have pretty good ratios, but a lot of the ones with no protein or bad segmentation don’t. You could potentially try to come up with some filters to try to more clearly get the cells you wanted, though it might be tricky. Definitely it’s worth trying to clean the segmentation, though I’m not sure how much better it can get due to the clumping.

I think if you swap though from trying to look at the absolute average difference at the plasma membrane to looking for what percent of cells show at least a moderate plasma membrane enrichment, the differences might be more apparent; if I use a (somewhat arbitrary based on the image I showed) threshold of your “Ratio1” of 1.1 as a threshold for enrichment (after switching to measuring in Crop rather than Corr), I get 58.7% of cells showing PM enrichment in the first image and 16% in the second, which to me shows a real difference. I’ve attached the edited pipeline here. PM-cytoplasm_BC.cppipe (21.6 KB)

As for the Vitra, remember that Cell = Nucleus + Cytoplasm, so the best measure of translocation is going to be Nucleus/Cytoplasm (or Cyt/Nuc), not Nuclei/Cell or Cytoplasm/Cell. If you do that, just from the back of the envelope you posted here, you’d end up with (.197/.136)=1.45 for image 1 and (0.206/0.175)=1.18 in image 2, which is a much bigger change.


It’s good idea to quantify percent cells other than intensity. I will try that. Thanks a lot for helps!