Problem in filtering transfected/untransfected cells by intensity


I am currently during experiments where I am comparing transient transfected/untransfected NIH3T3 cells with various mutants using cellprofiler. The transfected cells are coloured green by an HA-tag in the GFP channel. Everything in the pipeline works fine except for this:

I am currently seperating the transfected and untransfected cells by the ‘FilterObjects’ module and then defining transfected cells as objects with a minimum mean HA-tag intensity of X.

The problem is that pictures from the same experiment are not correctly filtered by the same intensity threshold; if it’s a low minimum intensity value, it works fine in some pictures but includes too many clearly untransfected cells in other pictures. If it’s higher, the untransfected cells are filtered out in some pictures (as they should be), but so are clearly transfected cells in other pictures.

I have tried playing around with the correctillumination calculate/apply modules, rescale intensity and enhance/supress features modules but nothing has worked for me yet. It might be that I just don’t know how to apply them correctly

I have attached the pipeline as well as examples of pictures that don’t add up. For the first image (01) the minimum mean intensity need to be 1.5 to filter correctly, while the second image set (05) it needs the minimum to be at 1.2.
Pipeline JNO 2.6 changed for 20201109 2 with no background editing.cpproj (2.2 MB) NIH3T3 D1 RhoA E40Q 01-Image Export-08_b0c0x0-1388y0-1040.tif (4.1 MB) NIH3T3 D1 RhoA E40Q 01-Image Export-08_b0c1x0-1388y0-1040.tif (4.1 MB) NIH3T3 D1 RhoA E40Q 01-Image Export-08_b0c2x0-1388y0-1040.tif (4.1 MB)
NIH3T3 D1 RhoA E40Q 05-Image Export-12_b0c0x0-1388y0-1040.tif (4.1 MB) NIH3T3 D1 RhoA E40Q 05-Image Export-12_b0c1x0-1388y0-1040.tif (4.1 MB) NIH3T3 D1 RhoA E40Q 05-Image Export-12_b0c2x0-1388y0-1040.tif (4.1 MB)

Hi @EllenAppel,

I took a look at your pipeline. Nice work!

You are applying the FilterObjects module correctly, but I think it’s challenging to find a good threshold because the intensities vary between your images. As you noted, what works well for one image doesn’t work on the others. These intensities might reflect different transfection levels or just differences in acquisition.

Instead of using a MeanIntensity threshold for the HAtag signal, I had good luck in your sample images using the standard deviation of HAtag intensity values within the cells (select StdIntensity for the Measurement; I used a threshold of 0.005). The WorkspaceViewer is nice for visualizing the output of this type of analysis:

In doing so, the absolutely intensity values of each cell won’t matter for being selected as above threshold – rather you’re selecting for cells that have a wider range of pixel values. I think this works well for the images that you sent because many of them have a brighter perinuclear expression, which means that the cells have a higher std deviation of intensity values.

Bonus note: in the MeasureObjectIntensity module you can select multiple objects and multiple objects to measure in a single module rather than duplicating the module. This can help streamline your analysis by reducing extra modules (but the way you’ve done it works too!).


1 Like

Pearl thank you so so much!

I have run through a lot of my other pictures and it works like a charm! I ended up changing it to 0.006 but I will definitely be using the standard deviation intensity filter from now on. I still have some other minor problems, but this was a huge step for the work on my pipeline :smiley:

1 Like