Colocalisation

[Paul Buchanan]

Hi All,

Hoping you can give me a few pointers with regards to co-localisation in image j and FiJi.

Firstly can some one explain to me in a bit more about intensity based co-localisation. To generate a histogram from which the correlation co-efficient can be generated, the intensity of one can is plotted against the other. But how does this take into account the location of the pixels? Sure this method just tells you if pixels in one channel have a similar intensity to another channel? How does that describe co-localisation. I have read all the papers referenced for the plug ins but just missing out on a few basic principles.

I have been using coloc2 and JaCop to look at receptor co-localisation with various endsomes as it traffics through the cell. Firstly i have been finding JaCoP extremely slow, so slow i cant run it with my images. Ok my images are ~200mb each and have about 20 slices in each but coloc2 runs through in about 5 - 10 minutes. I have been getting some r values and they seem to represent what i see in the overlay images with regards to more or less yellow colour. But i have been asked when explaining my work what exactly the Pearsons coefficient shows and as stated above am unsure if it represents location of co-localisation as well as the intensity. Also am not sure if this r value represents the whole z stack i have imaged? and what what about slices that have no co-localisation due to being above or below the cell does this on skew the results.

I have been looking into doing some object based colocalisation as it seems like it would be a better method to confirm co localisation over my z stacks but again jacop takes ages to do anything.

Any advice would be gratefully appreciated.

Thanks

Paul Buchanan

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[sonia fargue]

Hi Paul,

I will not comment on principles of colocalisation, as I not competent
enough and there better people to answer that.
I just wanted to point out that one reason you may find the JACoP so slow
is that you may be running too many analyses in it. I have found that by
selecting which parameters I really needed, there was little difference
between the two.
Depending on your data, the JACoP also offers the possibility to adjust the
thresholds manually, which in some situations may be more important (if
matching intensities in the pixels where correlation is analysed are not of
primary interest, but objects are).

Again, there are lots of knowledgeable people in the community.
For a good overview of colocalisation analysis principles, I found the
review by Bolte and Cordelière (J Microscopy 2006) most helpful. You may
already have read it, in which case good luck!


regards
Sonia Fargue


2013/9/19 Paul Buchanan <[hidden email]>

> Hi All,
>
> Hoping you can give me a few pointers with regards to co-localisation in
> image j and FiJi.
>
> Firstly can some one explain to me in a bit more about intensity based
> co-localisation. To generate a histogram from which the correlation
> co-efficient can be generated, the intensity of one can is plotted against
> the other. But how does this take into account the location of the pixels?
> Sure this method just tells you if pixels in one channel have a similar
> intensity to another channel? How does that describe co-localisation. I
> have read all the papers referenced for the plug ins but just missing out
> on a few basic principles.
>
> I have been using coloc2 and JaCop to look at receptor co-localisation
> with various endsomes as it traffics through the cell. Firstly i have been
> finding JaCoP extremely slow, so slow i cant run it with my images. Ok my
> images are ~200mb each and have about 20 slices in each but coloc2 runs
> through in about 5 - 10 minutes. I have been getting some r values and they
> seem to represent what i see in the overlay images with regards to more or
> less yellow colour. But i have been asked when explaining my work what
> exactly the Pearsons coefficient shows and as stated above am unsure if it
> represents location of co-localisation as well as the intensity. Also am
> not sure if this r value represents the whole z stack i have imaged? and
> what what about slices that have no co-localisation due to being above or
> below the cell does this on skew the results.
>
> I have been looking into doing some object based colocalisation as it
> seems like it would be a better method to confirm co localisation over my z
> stacks but again jacop takes ages to do anything.
>
> Any advice would be gratefully appreciated.
>
> Thanks
>
> Paul Buchanan
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[ctrueden]

Hi Paul,

> Hoping you can give me a few pointers with regards to co-localisation

Did you see the documentation at:
http://fiji.sc/Colocalization
?

It goes into great detail about the colocalization workflow, and includes
links to further reading about Pearson, Manders, Costes and Li methods.

Regards,
Curtis


On Thu, Sep 19, 2013 at 9:20 AM, Paul Buchanan <[hidden email]> wrote:

> Hi All,
>
> Hoping you can give me a few pointers with regards to co-localisation in
> image j and FiJi.
>
> Firstly can some one explain to me in a bit more about intensity based
> co-localisation. To generate a histogram from which the correlation
> co-efficient can be generated, the intensity of one can is plotted against
> the other. But how does this take into account the location of the pixels?
> Sure this method just tells you if pixels in one channel have a similar
> intensity to another channel? How does that describe co-localisation. I
> have read all the papers referenced for the plug ins but just missing out
> on a few basic principles.
>
> I have been using coloc2 and JaCop to look at receptor co-localisation
> with various endsomes as it traffics through the cell. Firstly i have been
> finding JaCoP extremely slow, so slow i cant run it with my images. Ok my
> images are ~200mb each and have about 20 slices in each but coloc2 runs
> through in about 5 - 10 minutes. I have been getting some r values and they
> seem to represent what i see in the overlay images with regards to more or
> less yellow colour. But i have been asked when explaining my work what
> exactly the Pearsons coefficient shows and as stated above am unsure if it
> represents location of co-localisation as well as the intensity. Also am
> not sure if this r value represents the whole z stack i have imaged? and
> what what about slices that have no co-localisation due to being above or
> below the cell does this on skew the results.
>
> I have been looking into doing some object based colocalisation as it
> seems like it would be a better method to confirm co localisation over my z
> stacks but again jacop takes ages to do anything.
>
> Any advice would be gratefully appreciated.
>
> Thanks
>
> Paul Buchanan
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Michael Schell-2]

If you are analyzing 16-bit stacks, the auto-threhold routine in JaCoP will be very slow.  If you are using auto-thresholding, try changing images to 8-bit first.

On Sep 19, 2013, at 11:10 AM, sonia fargue <[hidden email]> wrote:

> Hi Paul,
>
> I will not comment on principles of colocalisation, as I not competent
> enough and there better people to answer that.
> I just wanted to point out that one reason you may find the JACoP so slow
> is that you may be running too many analyses in it. I have found that by
> selecting which parameters I really needed, there was little difference
> between the two.
> Depending on your data, the JACoP also offers the possibility to adjust the
> thresholds manually, which in some situations may be more important (if
> matching intensities in the pixels where correlation is analysed are not of
> primary interest, but objects are).
>
> Again, there are lots of knowledgeable people in the community.
> For a good overview of colocalisation analysis principles, I found the
> review by Bolte and Cordelière (J Microscopy 2006) most helpful. You may
> already have read it, in which case good luck!
>
>
> regards
> Sonia Fargue
>
>
> 2013/9/19 Paul Buchanan <[hidden email]>
>
>> Hi All,
>>
>> Hoping you can give me a few pointers with regards to co-localisation in
>> image j and FiJi.
>>
>> Firstly can some one explain to me in a bit more about intensity based
>> co-localisation. To generate a histogram from which the correlation
>> co-efficient can be generated, the intensity of one can is plotted against
>> the other. But how does this take into account the location of the pixels?
>> Sure this method just tells you if pixels in one channel have a similar
>> intensity to another channel? How does that describe co-localisation. I
>> have read all the papers referenced for the plug ins but just missing out
>> on a few basic principles.
>>
>> I have been using coloc2 and JaCop to look at receptor co-localisation
>> with various endsomes as it traffics through the cell. Firstly i have been
>> finding JaCoP extremely slow, so slow i cant run it with my images. Ok my
>> images are ~200mb each and have about 20 slices in each but coloc2 runs
>> through in about 5 - 10 minutes. I have been getting some r values and they
>> seem to represent what i see in the overlay images with regards to more or
>> less yellow colour. But i have been asked when explaining my work what
>> exactly the Pearsons coefficient shows and as stated above am unsure if it
>> represents location of co-localisation as well as the intensity. Also am
>> not sure if this r value represents the whole z stack i have imaged? and
>> what what about slices that have no co-localisation due to being above or
>> below the cell does this on skew the results.
>>
>> I have been looking into doing some object based colocalisation as it
>> seems like it would be a better method to confirm co localisation over my z
>> stacks but again jacop takes ages to do anything.
>>
>> Any advice would be gratefully appreciated.
>>
>> Thanks
>>
>> Paul Buchanan
>>
>> --
>> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>>
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Paul Buchanan]

Thanks Sonia and Michael,

I will try converting the images to 8bit and see how that goes.

I was really only trying to use JaCoP for object based co localisation. Is this the best plug in for that or is there any better methods out there? My output from JaCoP for object based colocalisation only seems to give me coordinates and nothing to with colocalisation. Any thoughts why?

Also with JaCoP the thresholding tab, i dont need to change any of the settings, as it should do it automatically is that correct?

Also Sonia yeah i have read that paper but this not being my main area parts of it i didnt completely understand, hence why am still unsure of the process involved in intensity based colocalisation and how it takes into account pixel location and also if these plugins take into account the whole 3d stack?

Thanks again for your help.

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Paul Buchanan]

Hi Curtis

I have read that site you mentioned and most of the papers in the references. But just things havent quite clicked yet.

Do these plugins take into account the whole z stack? IF so how does it do this, the the intensitys from each stack just added to the overall histogram.

Also the histogram is a plot of pixel intensity's of one channel against another. How does this take into location? Am just confused about this histogram and if someone can give me a basic explanation of what exactly it plots

thanks

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Michael Schell-2]

> Also with JaCoP the thresholding tab, i dont need to change any of the settings, as it should do it automatically is that correct?
>

In theory, using  auto-threshold is a good thing because it avoids bias and/or error introduced by manual thresholding.  In practice, I find that the auto-threshold in JaCoP (Costes') sets the value too low compared to what my eye thinks is biologically relevant.   For a non-nuclear organelle, using cells stained with antibodies that produce low background, I tend to set the manual threshold to the first value above when all pixels in the nuciei are below threshold.  This value is always higher than the Costes' auto-threshold.  The biological question being tested should be considered when when sets a manual threshold.

Michael
--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Paul Buchanan]

Is there a way to do the thresholding using another method and then use that with JaCoP?

I also noticed that costes sets the threshold to low.

Surely setting by hand isnt consistent if your analysising a lot of images?

What did you mean by "when all pixels in the nuciei are below threshold", what does that look like i.e. the pixels in the nuclei you cant see?

Thanks for your help again

Paul

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html