Re: Colocalization analysis w/ ICA and JACoP - No intensity saturation please

[Daniel James White]

Hi Bryan,

Looking at cell1_Green.tiff and the red one that goers with it,
it is clear that the images are very badly over exposed / clipped  / saturated.

in cell2 the green is not saturated, but there is a saturated pixel in the red channel.
(its significant as the image is small)

cell3 is badly saturated.

This is very very bad news in general for quantitative analysis and especially for coloc analysis.
Yuo cant use images like this in this kinf of analysis.
You will get false/meaningless results.

Essentially, you do not know the true intensity of the brightest pixels,
since their intensities were chpoosed off at the top of the detectors range.
You lost the info that you are most interested in.
You must collect image data with in the range of the detectors ability.

You must use a HiLo or similar range indicator look up table / pallette
to set the imaging parameters of the detector  so the background is just above zero,
and the brightest pixels are not at the top of the range (typically 255 or 4095 etc. )

When I see no saturated pixels - i really mean NONE.

Also the offset is set a bit high: there should be very few or no 0 pixels,
to make sure you dont lost the info at the lowe end of the scale also.

You also need good signal:noise for these method,
as the methods used to measure coloc are very sensitive to noise.

For a tutorial on how to get it right look here:
http://pacific.mpi-cbg.de/wiki/index.php/Colocalization_Analysis

sorry for the bad news,
but I prefer to point you in the right direction than let you get it wrong.
I don't mean to cause embarrassment.
By no means are you the only person who ever made these mistakes, nearly everyone does.

cheers

Dan





On Jun 3, 2010, at 6:00 AM, IMAGEJ automatic digest system wrote:

>
> Date:    Wed, 2 Jun 2010 11:15:16 -0700
> From:    sibert <[hidden email]>
> Subject: Re: Colocalization analysis w/ ICA and JACoP
>
> Thanks,
>
> Originally I was just playing around with different plugins using whole
> images, narrowing the region of interest significantly narrowed the
> differences reported, especially within the Mander's coefficients.  
>
> The Pearson's coefficients and ICQ are still different between the two
> plugins with and without thresholds.  Notably, the ICA plugin seems to apply
> thresholds to these values, while thresholds are not applied by JACoP.  
>
> I've attached a .pdf, summarizing the reports from each plugin for three
> different cells (yeast) and one image set where each channel is from a
> different cell (non-colocalized).  I've also attached the images
> individually.  I hope to collect images at higher magnification and
> resolution in z- series stacks soon.
>
> -Bryan Sibert
>
> http://imagej.588099.n2.nabble.com/file/n5131910/colocalization.pdf
> colocalization.pdf
> http://imagej.588099.n2.nabble.com/file/n5131910/cell1_green.tif
> cell1_green.tif
> http://imagej.588099.n2.nabble.com/file/n5131910/cell1_red.tif cell1_red.tif
> http://imagej.588099.n2.nabble.com/file/n5131910/cell2_green.tif
> cell2_green.tif
> http://imagej.588099.n2.nabble.com/file/n5131910/cell2_red.tif cell2_red.tif
> http://imagej.588099.n2.nabble.com/file/n5131910/cell3_green.tif
> cell3_green.tif
> http://imagej.588099.n2.nabble.com/file/n5131910/cell3_red.tif cell3_red.tif
> http://imagej.588099.n2.nabble.com/file/n5131910/differentcells_green.tif
> differentcells_green.tif
> http://imagej.588099.n2.nabble.com/file/n5131910/differentcells_red.tif
> differentcells_red.tif
> --
> View this message in context: http://imagej.588099.n2.nabble.com/Colocalization-analysis-w-ICA-and-JACoP-tp5126856p5131910.html

Dr. Daniel James White BSc. (Hons.) PhD
Senior Microscopist / Image Visualisation, Processing and Analysis
Light Microscopy and Image Processing Facilities
Max Planck Institute of Molecular Cell Biology and Genetics
Pfotenhauerstrasse 108
01307 DRESDEN
Germany

[phone] (German Mobile)
[phone] (Work phone at MPI-CBG)
[phone] (Fax MPI-CBG LMF)

http://www.bioimagexd.net  BioImageXD
http://pacific.mpi-cbg.de                Fiji -  is just ImageJ (Batteries Included)
http://www.chalkie.org.uk                Dan's Homepages
https://ifn.mpi-cbg.de  Dresden Imaging Facility Network
dan (at) chalkie.org.uk
( white (at) mpi-cbg.de )

[sibert]

Hello,

Daniel James White wrote

Looking at cell1_Green.tiff and the red one that goers with it,
it is clear that the images are very badly over exposed / clipped  / saturated.

in cell2 the green is not saturated, but there is a saturated pixel in the red channel.
(its significant as the image is small)

cell3 is badly saturated.

This is very very bad news in general for quantitative analysis and especially for coloc analysis.
Yuo cant use images like this in this kinf of analysis.
You will get false/meaningless results.

Certainly these images could not be quantified, they were not taken with such intent.  At this point they are the only ones I had on hand to begin testing methods of analysis.  I wanted to make sure I could analyze the images properly before collecting them.

I did not think poor image quality would result in differences between quantification methods, but perhaps they handle saturated and empty pixels differently?  This would be important to know.

I've found a few sample colocalization image sets online and I have begun using those to test various methods of analysis.  It appears that there are still some differences between the ICA and JACoP plugins, although primarily in how they apply thresholds.  I would like to understand better what is going on; I think this has important implications for interpreting the results.  For example, Li et al. emphasize that 0-0 pixel pairs should not be counted in ICQ analysis.  The ICA documentation states that the lower threshold must be >0 and the upper threshold must be set to 255, but I do not know how JACoP addresses this.

Daniel James White wrote

For a tutorial on how to get it right look here:
http://pacific.mpi-cbg.de/wiki/index.php/Colocalization_Analysis

I appreciate the link, I have also found the following article to be very helpful in describing imaging requirements and various methods of quantification.

http://labs.pbrc.edu/cellbiology/documents/Colocalizationtutorial.pdf

-Thanks,

Bryan

[Daniel James White]

Hi, Bryan

Begin forwarded message:

>
> Certainly these images could not be quantified, they were not taken with
> such intent.  At this point they are the only ones I had on hand to begin
> testing methods of analysis.  I wanted to make sure I could analyze the
> images properly before collecting them.

Phew, thats ok then!


>
> I did not think poor image quality would result in differences between
> quantification methods, but perhaps they handle saturated and empty pixels
> differently?  This would be important to know.

For sure you must avoid:
1) detector Saturation
2) incorrect zero offset
3) noise

all three of these defeat the methods of the coloc plugins/methods.
1 and 2 make the intensity info non linear vs signal in the sample
3 makes the coloclaisation look weaker than it really is,
since it decreases the correlation between the 2 signals.

>
> I've found a few sample colocalization image sets online and I have begun
> using those to test various methods of analysis.  It appears that there are
> still some differences between the ICA and JACoP plugins, although primarily
> in how they apply thresholds.  

its likely (and I need to investigate further)
that these plugins implement the Costes auto threshold method in different ways.
Specifically how the implementation does the search for the thresholds below whioch there is 0 pearsons correlation.
You can start high and move down incrementally until you hit the first place where pearsons is 0 below thresholds,
or you could try to get there faster using a bisection or other minimization method to find it.
These might give slightly different thresholds on the same image, and thus different thresholded Pearsons and Manders numbers (and ICQ , is that thresholded??)

I plan to fing a fast robust way of doing the autothreshold and using that in the next generation coloc plugin we are designing.

> I would like to understand better what is
> going on;

me too.

> I think this has important implications for interpreting the
> results.  For example, Li et al. emphasize that 0-0 pixel pairs should not
> be counted in ICQ analysis.  

yes, and in the coloc plugins in Fiji, you can toggle on and off to include zero zero pixels.
It makes sense to ignore them if the image has many, as they falsely increase the correlation of the whole image.
thats why the offset zero value should be set do there are only a few if any zero values in the image.
If a alge area of the image is zero zero, then you definately need to use a region of interest to measure only where there is something to look at.
Better still take images with very few zeros - but withoiut a large positive offset either (don't waste dynamic range).

> The ICA documentation states that the lower
> threshold must be >0 and the upper threshold must be set to 255, but I do
> not know how JACoP addresses this.

Maybe Fabrice can fill us in?


Daniel James White wrote:
>
>
> For a tutorial on how to get it right look here:
> http://pacific.mpi-cbg.de/wiki/index.php/Colocalization_Analysis
>
>

I appreciate the link, I have also found the following article to be very
helpful in describing imaging requirements and various methods of
quantification.

http://labs.pbrc.edu/cellbiology/documents/Colocalizationtutorial.pdf

Yes thats the article that goes with JACoP plugin.

I hope to work together with those authors to design the next generation colc tool for ImageJ2.


cheers

Dan



Dr. Daniel James White BSc. (Hons.) PhD
Senior Microscopist / Image Visualisation, Processing and Analysis
Light Microscopy and Image Processing Facilities
Max Planck Institute of Molecular Cell Biology and Genetics
Pfotenhauerstrasse 108
01307 DRESDEN
Germany

[phone] (German Mobile)
[phone] (Work phone at MPI-CBG)
[phone] (Fax MPI-CBG LMF)

http://www.bioimagexd.net  BioImageXD
http://pacific.mpi-cbg.de                Fiji -  is just ImageJ (Batteries Included)
http://www.chalkie.org.uk                Dan's Homepages
https://ifn.mpi-cbg.de  Dresden Imaging Facility Network
dan (at) chalkie.org.uk
( white (at) mpi-cbg.de )