Li's Colocalization Analysis

[Daniel James White]

Hi Alexandra,

On Dec 10, 2011, at 6:00 AM, IMAGEJ automatic digest system wrote:

>
> Date:    Fri, 9 Dec 2011 10:55:51 -0500
> From:    Alexandra Sokolovski <[hidden email]>
> Subject: Li's Colocalization Analysis
>
> Hi,
>
> I have been experimenting with the different programs for Li's ICQ method
> to optimize my analysis. I find that the ICA plugin for ImageJ gives ICQ
> values that do not correspond to the PDM image when I select thresholding.
> I have used a control with staining of the same protein using two different
> fluorophores (GFP and antibody) which appears to give all positive PDMs
> based on the PDM image however the ICQ is very low (ICQ of -0.4 with
> thresholding and 0.4 without).

I can't comment on if that older implementation of Li has bugs or not....

>
> In an attempt to avoid this problem I switched to Fiji's Coloc2, which
> appeared to fixed this problem for the control. However based on figures
> that should have low ICQs they seem to be inflated. When I select smaller
> ROIs the ICQ goes down, so it seems like zero-zero pixels are being taken
> into account despite thresholding? I use cells so it is difficult to avoid
> zero-zero pixels as we move up the focal plane there is an increase in the
> number of zero-zero pixels. I tried solving this by using a mask, but
> values are lower than would be expected (ICQ = 0.2 for control used above).
>
> I was wondering if anyone has encountered this problem or knows of a
> solution.

It may or may not be a bug.
We eend to write more unit tests for Li method in Coloc_2
to see if the implementation we did gives the expected results.

We would appreciate your help here.
Maybe you can give us test images where you have an expected result or difference.

One ting is for sure, you must limit the analysis the the biological unit of interest,
eg a cell or cells, or some compartment or other.
Analysing the whole images only makes sense when there is no background
or non interesting areas.

please see docs and ideas here
http://fiji.sc/Colocalization_Analysis
http://fiji.sc/Colocalization_-_hardware_setup_and_image_acquisition
and
https://docs.google.com/document/d/1bEJyXdGyKx4J_G6WlsABpcl3cn-kI5CQs57qU-3Gx7Y/edit

cheers

Dan (and  Tom)


>
> Much appreciated,
>
> Alexandra Sokolovski

Dr. Daniel James White BSc. (Hons.) PhD

Leader - Image Processing Facility,
Senior Microscopist,
Light Microscopy Facility.

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)
chalkie666 Skype
http://www.bioimagexd.net  BioImageXD
http://fiji.sc                                        Fiji -  is just ImageJ (Batteries Included)
http://www.chalkie.org.uk                Dan's Homepages
https://ifn.mpi-cbg.de  Biopolis Dresden Imaging Platform (BioDIP)
dan (at) chalkie.org.uk
( white (at) mpi-cbg.de )

[alexa_11]

On Sat, Dec 10, 2011 at 5:46 AM, daniel white <[hidden email]> wrote:

> Hi Alexandra,
>
> On Dec 10, 2011, at 6:00 AM, IMAGEJ automatic digest system wrote:
>
> >
> > Date:    Fri, 9 Dec 2011 10:55:51 -0500
> > From:    Alexandra Sokolovski <[hidden email]>
> > Subject: Li's Colocalization Analysis
> >
> > Hi,
> >
> > I have been experimenting with the different programs for Li's ICQ method
> > to optimize my analysis. I find that the ICA plugin for ImageJ gives ICQ
> > values that do not correspond to the PDM image when I select
> thresholding.
> > I have used a control with staining of the same protein using two
> different
> > fluorophores (GFP and antibody) which appears to give all positive PDMs
> > based on the PDM image however the ICQ is very low (ICQ of -0.4 with
> > thresholding and 0.4 without).
>
> I can't comment on if that older implementation of Li has bugs or not....
>
> >
> > In an attempt to avoid this problem I switched to Fiji's Coloc2, which
> > appeared to fixed this problem for the control. However based on figures
> > that should have low ICQs they seem to be inflated. When I select smaller
> > ROIs the ICQ goes down, so it seems like zero-zero pixels are being taken
> > into account despite thresholding? I use cells so it is difficult to
> avoid
> > zero-zero pixels as we move up the focal plane there is an increase in
> the
> > number of zero-zero pixels. I tried solving this by using a mask, but
> > values are lower than would be expected (ICQ = 0.2 for control used
> above).
> >
> > I was wondering if anyone has encountered this problem or knows of a
> > solution.
>
> It may or may not be a bug.
> We eend to write more unit tests for Li method in Coloc_2
> to see if the implementation we did gives the expected results.
>
> We would appreciate your help here.
> Maybe you can give us test images where you have an expected result or
> difference.
>
>

I have a couple of images that I would like to send, but I'm not sure how
to attach them as the files are quite large.


> One ting is for sure, you must limit the analysis the the biological unit
> of interest,
> eg a cell or cells, or some compartment or other.
> Analysing the whole images only makes sense when there is no background
> or non interesting areas.
>
>
Because I take images of stacks the only way to avoid zero-zero pixels is
to create a mask, especially when looking at dendrites with punctate
staining. The masking however gives different values depending on which
channel I use a mask for. The values given also seem too low for an image
that should have nearly perfect ICQ.

> please see docs and ideas here
> http://fiji.sc/Colocalization_Analysis
> http://fiji.sc/Colocalization_-_hardware_setup_and_image_acquisition
> and
>
> https://docs.google.com/document/d/1bEJyXdGyKx4J_G6WlsABpcl3cn-kI5CQs57qU-3Gx7Y/edit
>
> cheers
>
> Dan (and  Tom)
>
>
> >
> > Much appreciated,
> >
> > Alexandra Sokolovski
>
> Dr. Daniel James White BSc. (Hons.) PhD
>
> Leader - Image Processing Facility,
> Senior Microscopist,
> Light Microscopy Facility.
>
> 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)
> chalkie666                                      Skype
> http://www.bioimagexd.net       BioImageXD
> http://fiji.sc                                  Fiji -  is just ImageJ
> (Batteries Included)
> http://www.chalkie.org.uk               Dan's Homepages
> https://ifn.mpi-cbg.de                  Biopolis Dresden Imaging Platform
> (BioDIP)
> dan (at) chalkie.org.uk
> ( white (at) mpi-cbg.de )
>
>

[dscho]

Hi Alexandra,

On Sun, 11 Dec 2011, Alexandra Sokolovski wrote:

> On Sat, Dec 10, 2011 at 5:46 AM, daniel white <[hidden email]> wrote:
>
> I have a couple of images that I would like to send, but I'm not sure
> how to attach them as the files are quite large.

Use Fiji's Help>Upload Sample Image to upload them and notify Dan, he has
access to the uploaded data.

Ciao,
Johannes

[Daniel James White]

On Dec 12, 2011, at 6:00 AM, IMAGEJ automatic digest system wrote:

>>>
>>> In an attempt to avoid this problem I switched to Fiji's Coloc2, which
>>> appeared to fixed this problem for the control. However based on figures
>>> that should have low ICQs they seem to be inflated. When I select smaller
>>> ROIs the ICQ goes down, so it seems like zero-zero pixels are being taken
>>> into account despite thresholding? I use cells so it is difficult to
>> avoid
>>> zero-zero pixels as we move up the focal plane there is an increase in
>> the
>>> number of zero-zero pixels. I tried solving this by using a mask, but
>>> values are lower than would be expected (ICQ = 0.2 for control used
>> above).
>>>
>>> I was wondering if anyone has encountered this problem or knows of a
>>> solution.
>>
>> It may or may not be a bug.
>> We eend to write more unit tests for Li method in Coloc_2
>> to see if the implementation we did gives the expected results.
>>
>> We would appreciate your help here.
>> Maybe you can give us test images where you have an expected result or
>> difference.
>>
>>
> I have a couple of images that I would like to send, but I'm not sure how
> to attach them as the files are quite large.

you can get a dropbox account and put them there and share them with me
or use the Fiji menu item
Help - upload sample image
then let me know

>> One ting is for sure, you must limit the analysis the the biological unit
>> of interest,
>> eg a cell or cells, or some compartment or other.
>> Analysing the whole images only makes sense when there is no background
>> or non interesting areas.
>>
> Because I take images of stacks the only way to avoid zero-zero pixels is
> to create a mask, especially when looking at dendrites with punctate
> staining.

To create the right make, you need to define very carefully what is the biological unit of interest,
and develop a way to segment that from the image = create the mask.

you might need another marker that shows you, eg, the extent of the cdll, or the cytoplasm,
or nucleus, or whatever it is that you want to measure coloc "in".

this is a hard problem.

> The masking however gives different values depending on which
> channel I use a mask for. The values given also seem too low for an image
> that should have nearly perfect ICQ.

Dont trust your eyes and what you "see" in the image.
Your eyes only see the really obvious things, and can easily be mislead from what is really there.

Thats why you are using a hopefully robust objective measurement.

cheers

Dan

>
>> please see docs and ideas here
>> http://fiji.sc/Colocalization_Analysis
>> http://fiji.sc/Colocalization_-_hardware_setup_and_image_acquisition
>> and
>>
>> https://docs.google.com/document/d/1bEJyXdGyKx4J_G6WlsABpcl3cn-kI5CQs57qU-3Gx7Y/edit
>>
>> cheers
>>
>> Dan (and  Tom)
>>
>>
>>>
>>> Much appreciated,
>>>
>>> Alexandra Sokolovski

Dr. Daniel James White BSc. (Hons.) PhD

Leader - Image Processing Facility,
Senior Microscopist,
Light Microscopy Facility.

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)
chalkie666 Skype
http://www.bioimagexd.net  BioImageXD
http://fiji.sc                                        Fiji -  is just ImageJ (Batteries Included)
http://www.chalkie.org.uk                Dan's Homepages
https://ifn.mpi-cbg.de  Biopolis Dresden Imaging Platform (BioDIP)
dan (at) chalkie.org.uk
( white (at) mpi-cbg.de )

[alexa_11]

>
>
> > I tried creating a mask by merging both channels and it seems to greatly
> > improve the ICQ outcome (ICQ = 0.344). If you have a dropbox account I
> > can send you a link to the images.
>
> Good that merging the channels improved the result. You might even want
> to threshold the merged images a bit, depending on the background noise.
>
>

I ran channel 1 through coloc2 twice and the math seems to be consistent
with Li's paper as I end up with an ICQ of 0.5.

> I will send you a dropbox account off-list and would be happy to look at
> the images and see the differences by myself.
>
> >     For testing purposes I modified Li's ICQ calculation to ignore
> zero-zero
> >     pixels. You can get this either by cloning and compiling that branch:
> >
> >            $ git clone [hidden email]:tomka/fiji.git
> >            $ cd fiji
> >            $ git checkout coloc-2-li
> >            $ sh Build.sh
> >            $ ./fiji
> >
> >     or you can get the compiled version and replace the
> >     Colocalisation_Anaysis.jar file in your Fiji.app/plugins/ folder.
> (You
> >     might backup the original one.) I just uploaded it temporary [4].
> >
> >     When you now use the modified version, does it change the values
> >     significantly?
> >
> >
> > I tried substituting [4] for the original colocalization plugin. It is
> > still taking the zero-zero pixels into account, I have attached the PDFs
> > that I acquired by selecting a small ROI around the cell ("smaller ROI"
> > file) and then repeated by selecting a larger ROI around the cell
> > ("larger ROI"). The sign for the numbers was reversed, but other than
> > that the values were still being skewed by zeros.
>
> Why do you suspect that zero-zero pixels still get considered? If it is
> because of the "% zero-zero pixels information" it is kind of a
> misconception. I just modified the Li algorithm in the code to ignore
> zero-zero pixels. Unfortunately, this is not respected in the statistic
> just mentioned. Actually, it is only respected in Li's ICQ value, even
> the two Li histograms still respect the zero-zero pixels. Would it help
> to resolve the issue if I make the whole plugin ignore zero-zero pixels?
>
> And do I get you right that the my change made the sign of the ICQ value
> to change? If so it would actually be a big difference and we should
> think about having an "exclude zero-zero pixels" option, shouldn't we?
>
> Cheers,
> Tom
>
>