Colocalization analysis w/ ICA and JACoP

[sibert]

Hello,

I would like to analyze the colocalization of two proteins in some fluorescent images.  Based on what I've read, I'd like to use the ICA method by Li et al. (J. Neurosci. 24:4070, 2004)

So far, I've tried the Intensity Correlation Analysis plugin and JACoP.

These plugins always provide different values for the Intensity Correlation Quotient (ICQ) as well as the Pearson's coefficient and Mander's coefficients for the same images.  I've tried both plugins with identical thresholds and without thresholds.  Adjusting the threshold does not appear to have any effect on the ICQ from JACoP.

The ICQ from JACoP seems to be excessively high, even when minimal colocalization is visible the value given is generally > 0.3.  Whereas the ICA plugin returns values of ~.28 when the proteins appear highly colocalized (such an image returns ~.44 in JACoP).

The Pearson's and Mander's coefficients are also generally higher in JACoP than ICA.

Unfortunately, I don't know Java well enough to determine how these calculations are being done in the source code.  It isn't clear to me from any documentation why the values are not identical between the two plugins.

Can anyone provide some insight as to why these plugins are providing different values for the same analysis and how I may discern which of the plugins is providing accurate calculations?

[Daniel James White]

Hi there "Sibert"

On Jun 2, 2010, at 6:00 AM, IMAGEJ automatic digest system wrote:

>
> Date:    Tue, 1 Jun 2010 09:28:55 -0700
> From:    sibert <[hidden email]>
> Subject: Colocalization analysis w/ ICA and JACoP
>
> Hello,
>
> I would like to analyze the colocalization of two proteins in some
> fluorescent images.  Based on what I've read, I'd like to use the ICA method
> by Li et al. (J. Neurosci. 24:4070, 2004)

it seems like a good method,
but you might as well also give the Manders (thresholded) and Pearsons (total and thresholded) measurements.

Remember to use a region of interest where the biology is, not the whole image.

>
> So far, I've tried the
> http://www.uhnresearch.ca/facilities/wcif/software/Plugins/ICA.html
> Intensity Correlation Analysis plugin  and
> http://rsbweb.nih.gov/ij/plugins/track/jacop.html JACoP .
>
> These plugins always provide different values for the Intensity Correlation
> Quotient (ICQ) as well as the Pearson's coefficient and Mander's
> coefficients for the same images.  I've tried both plugins with identical
> thresholds and without thresholds.  Adjusting the threshold does not appear
> to have any effect on the ICQ from JACoP.

Not sure if thresholds are used in ICQ....

but i am looking very hard at all these methods at the moment,
with the aim of making the next generation colocalization tool for ImageJ2...

I hope to do this in collaboration with Fabrice C and others who have contributed the current plugins.
Maybe Fabrice and comment on the difference between the results of the two plugins?
For sure we need to find out why they are different and implement it correctly in any new plugin.

>
> The ICQ from JACoP seems to be excessively high, even when minimal
> colocalization is visible the value given is generally > 0.3.  Whereas the
> ICA plugin returns values of ~.28 when the proteins appear highly
> colocalized (such an image returns ~.44 in JACoP).
>

it might be due to differences in the precision of the numbers used in the calculations,
but i expect that would not give differences this large.

Please can you send me a sample image that gives you different ICQs in the 2 plugins so i can verify the problem?

> The Pearson's and Mander's coefficients are also generally higher in JACoP
> than ICA.

by how much? Third significant figure differing by 1 is ok... larger than that is a problem...

>
> Unfortunately, I don't know Java well enough to determine how these
> calculations are being done in the source code.  It isn't clear to me from
> any documentation why the values are not identical between the two plugins.

One of the missions i have set myself is to make the code understandable
for a non expert, so they can understand the mathds being done,
and verify it agains the description in the original publiucations....

lets say... thats not always easy with the current crop of plugins, and that can be improved.

>
> Can anyone provide some insight as to why these plugins are providing
> different values for the same analysis and how I may discern which of the
> plugins is providing accurate calculations?

if you send me a sample image, I will be happy to investigate,
since i need to check both implementations anyway....


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 )

[sibert]

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

colocalization.pdf
cell1_green.tif
cell1_red.tif
cell2_green.tif
cell2_red.tif
cell3_green.tif
cell3_red.tif
differentcells_green.tif
differentcells_red.tif

[Henry007]

Hi. Does anybody know how to pick up ROI while using JACOP?
Fiji has this function while applying Manders/Costes  measurements; however, I have not found it while using JACOP.

Best,
Henry

[Veronica Labrador Cantarero]

Hello everyone,

     I have also found differences in Pearson and ICQ coefficients between JACoP and
ICA plugins. Please, have a look at the document "JACoP vs ICA" where
the same image was used with different grades of overlap.

     In addition, I have found that Overlap coefficients are higher than expected,
in my opinion, after applying a threshold even if the degree of colocalization
is low.

     I also compared Overlap and K coefficients obtained for several images with MS
Excel and with JACoP plugin. When no threshold was applied the results were
exactly the same but as I applied thresholds values became different (see the
document "Test 2").



We would really appreciate any help on this subject.


Here is the link to download the documentation:

http://rapidshare.com/files/396731811/Documentation.zip.html



Thank you in advance,

Veronica Labrador Cantarero

Servicio de Microscopia Optica y Confocal
Centro de Biologia Molecular Severo Ochoa
C/Nicolas Cabrera,1
Universidad Autonoma de Madrid
Madrid (Spain)
http://www.cbm.uam.es/confocal

[Gabriel Landini]

On Tuesday 08 Jun 2010  09:02:21 you wrote:
>
> http://rapidshare.com/files/396731811/Documentation.zip.html

Just one observation on something mentioned in the pdf document. You should
not apply "a posteriori" background correction and expect sensible results
because that type of procedure changes the data that you are precisely trying
to extract.
"A posteriori" methods use data from the image, not from the imaging system
alone. The same ball rolling algorithm on another image will change the image
differently depending on the image contents, despite that the background
correction that needs to applied is always the same.
You need to apply a different type of background correction by characterising
your imaging setup.

G.

[Daniel James White]

Dear Veronica,


On Jun 9, 2010, at 6:00 AM, IMAGEJ automatic digest system wrote:

> Date:    Tue, 8 Jun 2010 12:30:41 -0400
> From:    Veronica Labrador Cantarero <[hidden email]>
> Subject: Re: Colocalization analysis w/ ICA and JACoP
>
> Hello everyone,
>
>     I have also found differences in Pearson and ICQ coefficients betwee=
> n JACoP and
> ICA plugins. Please, have a look at the document "JACoP vs ICA" where
> the same image was used with different grades of overlap.

thanks for sharing  that. It see what you mean.... i will also have  closer look.
It loks like you have done sensible tests there....
Maybe i can use something similar for unit testing of the next generation coloc tool we are designing.
Maybe you might like to also help with that?

>
>     In addition, I have found that Overlap coefficients are higher than =
> expected,
> in my opinion, after applying a threshold even if the degree of colocaliz=
> ation
> is low.

I cant see how to do thresholded ICQ in JACoP?
How did you do that?

>
>     I also compared Overlap and K coefficients obtained for several imag=
> es with MS
> Excel and with JACoP plugin. When no threshold was applied the results we=
> re
> exactly the same but as I applied thresholds values became different (see=
> the
> document "Test 2").

again, i dint know how JACoP can use thresholds for that?????

>
>
>
> We would really appreciate any help on this subject.=20

Well, I hope we can help each other a a community and find the problems,
and work on more robust tools for the future.
JACoP is a big step forward in usability from the older stand alone plugins,
and we need to work hard to make the next generation toold even better for imageJ2

>
>
> Here is the link to download the documentation:
>
> http://rapidshare.com/files/396731811/Documentation.zip.html

many thanks for sharing, its really helpful!

cheers

Dan

>
>
>
> Thank you in advance,
>
> Veronica Labrador Cantarero
>
> Servicio de Microscopia Optica y Confocal=20
> Centro de Biologia Molecular Severo Ochoa
> C/Nicolas Cabrera,1
> Universidad Autonoma de Madrid
> Madrid (Spain)
> http://www.cbm.uam.es/confocal

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]

Veronica Labrador Cantarero wrote

Hello everyone,

     I have also found differences in Pearson and ICQ coefficients between JACoP and
ICA plugins. Please, have a look at the document "JACoP vs ICA" where
the same image was used with different grades of overlap.

I'm in the process of trying to expand my knowledge of Java and go through the ICA and JACoP plugins.  At this point, I've only been able to find the source for JACoP v. 1, not the current version, but the changelog does not indicate any changes to the ICQ calculation.

The following are some differences I believe exist:

-ICA allows you to place a threshold on ICQ, JACoP does not.
-ICA does not include pixels less than one when calculating the mean for each channel (n = total pixels - zero pixels), while JACoP does (n = total pixels).
-ICA does not include zero-zero pairs in calculation of the ICQ ((positive ICQ pairs)/(total pairs - zero/zero pairs)), JACoP does (positive ICQ pairs/total pairs)
-JACoP normalizes the intensity range of each pixel to 0-1 ((A[i] - Amin)/Amax) prior to calculating ICQ value for each pair , this will skew your data slightly if Amin is not 0.

Based on this, it should hold that an image with very few or no zero-zero pairs should provide the closest results between the two plugins, however because of the method of normalizing the data in JACoP there would still be a very slight difference (probably insignificant?).

I'm trying to write some new plugins to mimic the behavior of each plugin to confirm that I am correct.  I'm also trying to check for differences in the other values (Pearson's, Mander's).  It's possible that I have missed something in the code and these differences are incorrect or not comprehensive.

I will continue my efforts to nail this down, I welcome any confirmation/correction by others.

-Bryan Sibert

[sibert]

A few more notes on some differences between the ICA and JACoP plugins:

Pearson's coefficient

-The Pearson correlation for ICA does not include 0-0 pairs, JACoP does.

-A manual threshold can be applied to the Pearson correlation for ICA, only the automatic Costes' threshold can be applied for JACoP.

Mander's coefficients (M1 & M2)

-Thresholds in ICA are inclusive (>=), thresholds in JACoP are exclusive (>).

-ICA calculates M1 by calculating the sum of all ch1 pixels that overlap with ch2 pixels > ch2 threshold divided by the sum of all ch1 pixels; the minimum threshold ('use threshold' unchecked) is 1.

-JACoP v.2 calculates M1 by calculating the sum of ch1 pixels > ch1 threshold that overlap with ch2 pixels > ch2 threshold divided by the sum of ch1 pixels > ch1 threshold.

-JACoP v.1 calculates M1 similar to ICA.

Veronica Labrador Cantarero wrote

     I also compared Overlap and K coefficients obtained for several images with MS
Excel and with JACoP plugin. When no threshold was applied the results were
exactly the same but as I applied thresholds values became different (see the
document "Test 2").

Overlap Coefficients (Overlap, K1, & K2)

-When thresholds are applied JACoP will only include pixels that are above threshold for both channels.  This includes the numerator values (A*B) and the denominator values (A^2 and B^2).

-I have not observed any differences between the ICA and JACoP Overlap coefficient other than the difference in handling thresholds (inclusive vs. exclusive respectively) which is usually insignificant.

In the document 'Test 2' the denominator values contain pixels that are below threshold in the other channel, deleting these will give you the same values as the JACoP plugin.

[Daniel James White]

Dear Veronica and Bryan,

It really looks like we are gathering some useful information on all these plugins,
which is really important, since we are trying to design the next generation coloc tool for ImageJ2.

Thanks for sharing your observations!

Here are some I recently found out:

1) The Colocalization Threshold plugin uses a rearranged form of the Person's correlation equation,
the should be more numerically precise and faster to compute than the
way the equation is written in the classic Manders paper.
(Well done (Tony Collins?) whoever implemented that!)

2) I didn't find an implementation in java yet of the Costes statistical significance test (P-value)
(See Colocalization Test plugin and JACoP)
that follows the letter of the description of the algorithm in the Costes et al Paper.
It seems that the Coloc Test generates new random data that looks a bit like the image
but does not reshuffle the original image,
and the JACoP plugin does reshuffle the original image,
but it does it in a destructive way.
Maybe Fabrice can comment?

I really think we should pool all this information in a place where we can
all see it, contribute more and discuss the details.
This can be a place to pool info, and to add wishes to the list of potential features
for the next generation coloc tool for imageJ2

This list is great for sharing info of course,
but maybe we can also make a home for a coherent list of stuff we have found out.

... in fact thats easy with google docs...
and I will send you an invite to a google docs  document that have just started.
It would be great if you can copy paste the info you have already sent to this list
that you think fits.... there.....

Bryan, Please send me the email address you use on your google account
so i can invite you to also edit the document.
The others seemed to exits in google already.

The doc can also be viewed by anybody here:
http://docs.google.com/View?id=df66rgc7_2dtqkv3dx

more below....


On Jun 11, 2010, at 6:00 AM, IMAGEJ automatic digest system wrote:

>
> Date:    Thu, 10 Jun 2010 13:00:52 -0700
> From:    sibert <[hidden email]>
> Subject: Re: Colocalization analysis w/ ICA and JACoP

> I'm in the process of trying to expand my knowledge of Java and go through
> the ICA and JACoP plugins.  At this point, I've only been able to find the
> source for JACoP v. 1, not the current version, but the changelog does not
> indicate any changes to the ICQ calculation.

is the source code in the .jar you downloaded?
It was in the one i recently got!

just unzip the jar (rename it to .zip of you need)

>
>
> I'm trying to write some new plugins to mimic the behavior of each plugin to
> confirm that I am correct.  I'm also trying to check for differences in the
> other values (Pearson's, Mander's).  It's possible that I have missed
> something in the code and these differences are incorrect or not
> comprehensive.

You seem keep to write some code....
so we should work together!

>
> I will continue my efforts to nail this down, I welcome any
> confirmation/correction by others.

Lets put all this info, arguments, agreements, disagreements on the google docs document.
and use that asa forum for pinning things down, explaining what we think is the right way to do stuff,
and recording our wishes for the next generation plugin.

Many heads a re better then one!

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 )