Coloc2, Problem with Input Data, HELPPP

I’m trying to analyze colocalization between 2 signals. I would like to use Pearson coefficient.
That’s why I’m using coloc2 plugging.
For some images, it 's work without problem.
But in some case, I got a warning message :
Warning! Problem with input data - Pearson correlation: a numerical problem occurred: the input data is unsuitable for this algorithm. Possibily too few pixels.
And i don’t understand this message. Why I got this message for some images?? all images have the same size (pixel)… I draw a ROI
Despite this message, I optained a Pearson’s R value. Is it bad if i use it ???

If some can help me… I’m completely lost…
Thanks a lot

Dear @LeaA,

welcome to the forum and a happy new year. :slight_smile:

Are you using the same region of interest for all images or do they differ in size? If so, I’d assume that the size of the ROI is too small. In any case, could you provide sample images and details on how you have produced this error (see Bug reporting best practices for details on what might be helpful)?

In general, @chalkie666 and/or @etadobson might be able to help with the issue (I guess).


1 Like


@stelfrich is correct in all that he said. To start - just post some sample images and the details… then we can try to see if we can recreate your issue. We’ll do our best to help!

eta :slight_smile:


Too few pixels means the result will be statistically unreliable. Too few data points.
So you need more image data in the data set.
It can be a stack of images with a different roi per slice using the mask image to select where to analyse in each slice.

If the aitothresholds method also selects most of the region of interest is below zero correlation then there will be even fewer correlated pixels to measure.

Make sure the noise is low in the input data.
Deconvolve it or smooth it first to suppress photon shot and detector noises.


Does the plugin return a correlation coefficient in that case as well? If so, we could make the user-facing message more concise by saying that results produced might be statistically unreliable?

Thanks a lot for all your answers !!!
It is not the same ROI for all sample. I draw the shape of one cell for each sample. But the ROI is large as the cell takes half of the image.

I send you 2 samples:
Image1, i got warning message
Image2, everything is fine
I analyze the colocalization between the red and green signals. (The blue signal is the nucleus of the cell. )
I can not send you the .roi file. I’m sorry for that

Image 1

Image 2:

@LeaA for this image, number 1, the calculated ch2 threshold is down at 0 - which is not realistic, hence the warning and probably leads to problems. You might have a wrong offset setting on a confocal pmt, or subtracted too much background… (but there are 0% zero zero pixels… hmmm.
I get auto thresholds of 12 and 7 for the roi i used…

So it worked ok for me…
maybe look at your region of interest a bit more carefully.

1 Like

@LeaA … and i would not use the whole cell as the ROI… thge biology you care about is probably only happening in some subcompartment of the whole cell, eg not in the nucleus… but only in those large roundish objects???
Maybe just analyse the parts of the cell where the biology you care about is happening. Draw a mask image with the parts you want and use that for roi selection. Maybe use oneof the image channels to segment and make the mask image.

It looks like the pixels are way too small… 45 nm…
if you are using a high NA oil immersion lens, NA 1.4, then pixels should be 60-100 nm separated.

Your image is noisier than it could be because the signal is spread out over too many pixels. Empty magnification.

You need to deconvolve the images before analysis to suppress noise and fix the systematic error of the point spread function killing the contrast of small features.

At the very least you need to de noise the images, best by deconvolution, or at a push perhaps a gaussian or mean filter.

There is a scanner mis- calibration on your confocal, i can see you maybe used bidirectional scanning, but the adjavent lines are a little out of position with respect to each other… i can see the combing effect of that in the bright objeects.

you need to optimise the imaging conditions first to get good image date, then deconvolve it, then do this analysis on just the biologically interesting parts of the cell.
then you will get a more meaningful answer.

Dont forget to state the spatial image and optical resolution the experiment is done at - or else the results have no spatial size context.
“Pixel intensity correlations were measured at xxx resolution”


@stelfrich it still returns results values, unless the maths totally fails. A message as you suggest would be a good idea!