Colocalization analysis

Dear all,

I’m a bit troubled with analyzing some images.
I have to determine the amount of colocalization
of red and green in the following image.

To this end I’m using the Coloc2 Plugin (included in Fiji).
I calculated the approximate PSF for my images and
subtracted the background within Fiji (Subtract Background).
I’m not using a ROI yet but the whole image.

Now I’m getting warnings, that my results might not be reliable.
Should one deconvolve the image before?

Maybe you can give me any hints how to process the
example image provided below? Thanks in advance.
Example Image:

Settings:

 run("Coloc 2", "channel_1=[" + inputFiles[i] + " (red)] " + 
        "channel_2=[" + inputFiles[i] + " (green)]" +
       " ratio=50 threshold_channel_1=50 threshold_channel_2=50 display=255");

Best regards,
Stephan

Hi @stephanmg

Which warnings are you getting exactly?

Colocalization is a tricky analysis for sure - there are many pitfalls.

As stated on the Coloc 2 page on the ImageJ wiki“Deconvolution is a great way to restore images to give a better estimate of the real fluorophore spatial distribution by suppressing noise and removing the offset or background…” Manders/Pearsons are really sensitive to non-zero background - so applying a mask could also be helpful.

Try posting your image again when you can, providing exactly the same settings you used in your Coloc 2 run - so we can try to reproduce what you found.

eta

2 Likes

Thanks @etadobson.

I updated my post above.
Warnings are:

Zero-zero ratio too high
The ratio between zero-zero pixels and other pixels is large: 0.14. Maybe you should use a ROI.

y-intercept far from zero
The ratio of the y-intercept of the auto threshold regression line to the mean value of Channel 2 is high. This means the y-intercept is far from zero, implying a significant positive or negative zero offset in the image data intensities. Maybe you should use a ROI. Maybe do a background subtraction in both channels. Make sure you didn't clip off the low intensities to zero. This might not affect Pearson's correlation values very much, but might harm other results.

I’m now trying to apply a mask.

Best regards,
Stephan

@stephanmg

Just let us know if you want/need anything else…

eta

1 Like

Thanks @etadobson.

I’ll try to get rid of these warnings first and see what I get.

1 Like

A post was split to a new topic: Coloc2: calculating the approximate PSF

Hey all,

I’ve another question: Coloc2 does the “classical” colocalization with two colors, correct?
I’ve stumbled across the problem, that I have to do a color segmentation/feature extraction first,
in my images and then do a colocalization analysis.

I’m just wondering if it’s okay to first do a segmentation (Probably with some software X) and
then do the colocalization with e, g, Coloc2. Or if it is more appropriate to do this with one tool
which does both.

@All: This is an interesting read definitely if using Coloc2 is what you are doing: https://imagej.net/_images/e/e2/Costes_etalColoc.pdf

Thanks.

Best wishes,
Stephan

@stephanmg

No worries… So. I think you might have a typo in there - no? You want to look at 3 channels in total? The thing is… you can only compare (using classic colocalization measures - those in Coloc 2) two channels at a time. Can you describe in a bit more detail what you wish to do with your analysis and why?

eta

Sorry @etadobson.
I was a little bit short of time yesterday.

I edited my post above. It’s rather a conclusion than question.

Here comes the real question: I selected a ROI in the image I posted awhile ago (see above),
removed the background and split the channels into R, G, and B.
Still I’m getting a high correlation, where I’m visually
not seeing any (the amount of red and green pixels in the image above should be quantified):

!!! WARNINGS !!!
Warning! y-intercept far from zero - The ratio of the y-intercept of the auto threshold regression line to the mean value of Channel 2 is high. This means the y-intercept is far from zero, implying a significant positive or negative zero offset in the image data intensities. Maybe you should use a ROI. Maybe do a background subtraction in both channels. Make sure you didn't clip off the low intensities to zero. This might not affect Pearson's correlation values very much, but might harm other results.
Warning! Threshold of ch. 2 too high - Too few pixels are taken into account for above-threshold calculations. The threshold is above the channel's mean.
RESULTS:
Coloc_Test
% zero-zero pixels, 4.97
% saturated ch1 pixels, 0.00
% saturated ch2 pixels, 0.26
Channel 1 Max, 255.000
Channel 2 Max, 255.000
Channel 1 Min, 0.000
Channel 2 Min, 0.000
Channel 1 Mean, 1.724
Channel 2 Mean, 14.599
Channel 1 Integrated (Sum) Intensity, 10233194.000
Channel 2 Integrated (Sum) Intensity, 86631428.000
Mask Type Used, none
Mask ID Used, 196894066
m (slope), -278.89
b (y-intercept), 495.54
b to y-mean ratio, 33.94
Ch1 Max Threshold, 1.00
Ch2 Max Threshold, 255.00
Threshold regression, Costes
Pearson's R value (no threshold), -0.03
Pearson's R value (below threshold), -0.03
Pearson's R value (above threshold), -0.04
Li's ICQ value, 0.090
Spearman's rank correlation value, -0.03011675
Spearman's correlation t-statistic, -73.3977
t-statistic degrees of freedom, 5934094.000
Manders' M1 (Above zero intensity of Ch2), 0.676
Manders' M2 (Above zero intensity of Ch1), 0.678
Manders' tM1 (Above autothreshold of Ch2), 0.001
Manders' tM2 (Above autothreshold of Ch1), 0.678
Kendall's Tau-b rank correlation value, -0.0211
Costes P-Value, 0.00
Costes Shuffled Mean, 0.00
Costes Shuffled Std.D., 0.00
Ratio of rand. Pearsons >= actual Pearsons value , 1.00

Hope somebody has a suggestion.

Best wishes,
Stephan

@stephanmg - the Y-intercept warning above states:

so I would double-check your pixel values after background subtraction. Perhaps with the ROI selection you do - you don’t even need to do this background subtraction beforehand.

eta

Hey @etadobson. I checked the pixel values, which seem okay. I tried to do the same without background substraction. Still the second channel (green) gives me this warning.

Do you think that doing a Subtract Background with the Light Background option is better to use?

And more specifically, should one first split the channels and then do a background subtraction (Option A)? Or
should I first do a background subtraction and then split into channels (Option B)?

If I do a background subtraction with the light background option in ImageJ I get a different picture, see attached. (Maybe more revealing?)

Best, S.

@stephanmag
My advice is read the coloc2 documentation very carefully. Rest for a few hours before reading the Manders paper referenced there. Sleep then read the Costes paper referenced there also.
Understand the math then think about how you should apply it to your question. Maybe you want object based overlap method instead.

Next think hard about the biologically meaningful region of interest for the question. Think again about the question or hypothesis. Is there a mathematically definable hypothesis? Crystalise that in your mind
Then decide how much data you need to collect for good statistical power and remember a single absolute measurement is less informative than a comparison between a sample test data set vs positive and negative controls. Develop an assay then you have a way to test hypotheses

Hope it helps

4 Likes

@chalkie666 this helps indeed. will follow this procedure.

Thank you, I think I managed to do the Coloc now.
One question remains: Can I display somehow the pixels which have been colocalized with Coloc2?
I don’t see any option like this.

Best wishes,
Stephan