Colocalization analysis using Coloc2-help

Hello everyone,

I am trying to use FIJI Coloc2 plugin for my below 16-bit fluorescence images of 2 proteins (png files below). I am interested in calculating colocalization, especially at the clusters/puncta in 2 images.

In theory, these clustered proteins should co-localize well, resulting in high Pearson’s and Mander’s coefficients. I subtracted background and tried Coloc2 with below settings: (I calculated PSF using 1.22λ/NA formula to get the value of approx. 3)

I got below results:

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.
Pearson’s R value (no threshold): 0.54
Pearson’s R value (below threshold): 0.00
Pearson’s R value (above threshold): 0.31
Manders’ M1 (Above zero intensity of Ch2): 1.000
Manders’ M2 (Above zero intensity of Ch1): 0.982
Manders’ tM1 (Above autothreshold of Ch2): 0.773
Manders’ tM2 (Above autothreshold of Ch1): 0.591
Costes P-Value: 1.00

Can anyone figure out why I have the warning message though I subtracted background and selected ROI for calculation? And I am not sure if Pearson’s R value (above threshold): 0.31 is too low? I also tried using 8-bit binary mask but it was even worse.

Any help is much appreciated. Thank you in advance!


What are your calculated thresholds for ch1 and ch2? If at/near 0 - you might have a wrong offset setting on a confocal pmt, or subtracted too much background… Are these jpegs the datasets you are processing? Too - be sure to use the original file format acquired on your system. JPEGS ARE NOT APPROPRIATE for image analysis purposes (you can read more on that here).

How exactly are you selecting your ROIs? Where exactly is the biology you are interested happening within the cell? Have you tried drawing different ROIs and do you always get the same error?

If it’s a noise issue… then you might consider deconvolving your images first before coloc analysis. That might also help.

You can read through these older posts to find some helpful tips as well:


1 Like

Hi @etadobson, thanks for your reply. Here’s my response to your queries:

  1. Ch1 Max Threshold: 222.00, Ch2 Max Threshold: 599.00 (Note: these are TIRF images within 150 nm evanescent field)

  2. I analyzed TIFF images but I uploaded JPEGs. Here are the TIFF and PNG files (somehow TIFF files are not visible when uploaded).



  1. I drew the ROIs manually around the cells as below and analyzed each one separately:

However, I realized in the output pdf generated, it shows:
Mask Type Used: none
Mask ID Used: 1819056758
Does it mean that the ROI was not selected properly?

  1. The 2 proteins form discrete clusters and interact. So, the colocalization/interaction should occur within the clusters.

  2. Regarding noise, I think I have less noise at TIRF plane. Nonetheless, I can try deconvolving. Do you have a suggestion for a specific plugin/macro to try deconvolution?

  3. I have read the older posts you mentioned before I posted in the forum and tried the suggestions mentioned. However, I still ran into the issues.



Can you share your TIFF images via a link to a file-sharing service - like Dropbox or something? I want to take a look at them… your histograms, etc. Your reported thresholds are not near zero - so perhaps its more an issue of positive offset? As long as you don’t clip your low intensities to zero - should be fine subtracting background, etc. Though for sure - Deconvolution is the way to go.

For Deconvolution - there are a few different plugin options you can use… just click that links and check out what will work best for you and your data. If you have specific questions then for decon… just post another forum thread to get targeted help for that.

For the ROIs… just make sure you are saving your ROIs to the ROI Manager. And then in your Coloc 2 set-up - be sure that for the “ROI or mask” selection you choose “ROI Manager”. That should process them all at once, giving you results output per ROI.



Here’s the google drive link for the images:

I also realized there is a huge difference in Pearson’s and Mander’s values depending on whether I choose Costes or Bisection for threshold regression on selected ROI in background subtracted images.

I get higher positive coefficients when I used Bisection. So, does it mean Bisection threshold is better?


So - I think we are narrowing down the possible issues then…

Costes or Bissection implementation define the stepper used to move through the image - starting a SimpleStepper at max of channel 1 or a BissectionStepper at mid-point of channel 1, respectively. Bisection is a much faster way to find the correct thresholds because it needs fewer iterations to get to the same, hopefully, thresholds as Costes. Only noise would mess that up. Not much noise after proper deconvolution.

So again… try deconvolving your images first and then do the analysis once again - testing Costes versus Bissection still - then you should see similar thresholds.

As an aside - looking at your tiff images… there is significant offset in your images… non-zero background levels. When you were doing your background subtraction - how exactly did you do it and then what differences did you see in results before/after regarding Coloc? Decon should take care of this issue in any case though…



Regarding the background subtraction in previous results I uploaded, I selected a ROI outside the cell, measured the intensity and subtracted this value from the whole image.

Now, I have also tried rolling ball subtraction with different radius (10, 20 pixels). Below is the results I got:

I get higher positive coefficients with rolling ball background subtraction. Also, the y-intercept is closer to zero than in background ROI subtraction. So, is it better to use this?



Is it better?! Well - it depends… you can see how altering these background values changes your calculated coloc values as well. In general - I would say these background subtractions are necessary in your case - so be sure to document everything fully and it’s fine. You are doing anything ‘wrong’… but it also shows how important it is that images are first acquired using the full range of intensity values, etc.

You can also try a new, more robust colocalization method that was recently published. We have implemented this method via Ops. Here is a link to a script you can try running to test out this method on your data - at least you will get a p-value at the end to reveal the significance of those calculations (I’d recommend calculating the MTKT and Pearsons). Only if you want…


Thanks for sharing the new method. However, I am not familiar with Ops. Could you explain how am I supposed to implement and run this new method in FIJI, for eg: like a plugin/macro?

No worries.

ImageJ Ops are simply image processing algorithms. We’ve ported over a few colocalization algorithms… but what I mentioned/linked above is a new methodology that you can try out. To run the script - you should be able to simply copy/paste it into the Script Editor, select Language > Groovy, and then run it with two opened images.

But it’s just an alternative for your to test if you want…