Posted by
Daniel James White on
Dec 19, 2007; 8:59am
URL: http://imagej.273.s1.nabble.com/colocalization-greater-than-100-tp3697720p3697722.html
Hi Gretchen,
your method looks interesting,
but are there are several places where subjectivity creeps in?
To do pixel intensity spatial correlation coloc analysis
try the coloc plugins from macbiophotonics / imageJ plugins pages
First you must subtract background first - maybe use rolling ball or
subtract mean of a background area. You have to decise what is
background
and what isn't.
The auto threshold (costes method) in the
Colocalization Threshold plugin
will give you a reliable and repeatable non subjective threshold
bases on Pearsons correlation.
It will then calculate the Manders coefficients (amount of red coloc
with green and vice versa
Manders 1993 i think)
and show you the 2D histogram or scatter plot,
which is a good way to visualise the correlation between the 2 colour
channels.
I would publish the Manders coefficients, the thresholds that are
calculated, and the scatter plot (rather than a 2 colour merged image
- which is very difficult to interpret scientifically )
Then use the colocalisation test plugin,
with the Costes method to test the statistical significance of the
colo you see.
The coloc result are not solid without this test , as they could be
caused by random
overlap in a busy image.
Read the coloc docs at macbiophotonics:
http://www.macbiophotonics.ca/imagej/colour_analysis.htmOther things to consider to get good results.
1) you should image beads on the same system to see any spatial and
chromatic abberations in the spatial locations of the 2 colour
channels over the field of view.
You might be surprised how bad it is. (seems you are trying to do that
already)
2) on a zeiss confocal there are 3 emsission pinholes to adjust with
100 nm fluorescent beads
to get the signals coregistered. The engineeers only use the 10x lens
and large beads,
and always leave it in a non optimised state, no good for coloc of
small objects.
3) Olympus and leica confocals only have a single pinhole,
so als long as you are using a good (expensive) chromatically and
field corected lens
you are in good shape....
4) ... unless you are using a UV (say 405) laser coming through a
different optical fiber
into the system than the visible laser lines. The collimator must be
carefully adjusted to bring the
"blue" signal illuminated by the lioght from the UV fiber into
register with the visible illumination.
5) 2D images are ok for coloc, but real sampkles are 3D so in order to
get a good idea of the whole sample 3D imagging is a good thing.
6) If you are using a widefield system, consider deconvolution,
to increase contrast and remove out of focus fuzz,
but thats a whole kettle of fish in itself)
7) You might consider an object based coloc method,
depending on how your biology works.
Segment out objects, then see how much they overlap
between channels.
8) Bleed through of fluorescence from one channel into another is your
worst enemy in coloc.
You can see it clearly in the 2D histogram.
If you have bleed through, you will likely measure it as colocalisation,
which is in fact false. Watch out!
9) BioImageXD is also free like imageJ ,
and gives the same methods as in the plugins above,
but in a more coherent user interface.
If you have problems with that , send me some sample images,
and we will try to get it to work for you.
Say Hi to MN for me, I used to live in Rochester... Mayo Clinic...
cheers
Dan
Begin forwarded message:
> Date: Tue, 18 Dec 2007 22:18:48 -0600
> From: "Gretchen Unger, Ph.D." <
[hidden email]>
> Subject: colocalization greater than 100%
>
> Hi all,
>
> I used the division feature of Image J, v. 1.38 to estimate maximum
> colocalization between two optical sections, one was green (reporter
> gene) and the other red (cells). It looked like it worked but in one
> section I calculated a value over 100%. Can anyone explain how that
> might happen?
>
> I transformed final images processed against background into
> grayscale using Photoshop for the Image J calculations. In Image J, I
> calculated a joint (green+red) area by pixel-by-pixel division of
> green/red. This would calculate a maximum colocalization as a pixel
> with any signal above background in both images would divide to a
> "1". Because of drift or jumping or whatever, I had to manually alter
> alignment in some of the joint picture calculations.
>
> Next, I calculated "area fractions" in both the joint and red image
> using the "threshold" feature, applying threshold , and then
> "measuring" following selection of "area fraction" as an attribute.
> Because my final calculation was Colocalization =
> (green/red)*100/red, I selected the thresholding level from my
> denominator using the auto function. I then manually set the
> threshold to the same value (whatever it was) for the area fraction
> measurement of the joint image (the numerator). I got a number around
> 170% (17.25 divided by 10.37), I have used this method in the past
> and got numbers ranging from 50% to 98%, which seemed reasonable. I
> don't know why this set of aligned panels exceeds 100% and can't
> explain to anyone how that happened or what it can mean. Should I
> auto-threshold both images ? did I alter a calculation basis somehow
> when I manually aligned the image pair?
>
> Any thoughts or comments would be appreciated.
>
> Gretchen Unger
Dr. Daniel James White BSc. (Hons.) PhD
Senior Microscopist / Image Processing and Analysis
Light Microscopy Facility
Max Planck Institute of Molecular Cell Biology and Genetics
Pfotenhauerstrasse 108
01307 DRESDEN
Germany
New Mobile Number!!!
+49 (0)15114966933 (German Mobile)
+49 (0)351 210 2627 (Work phone at MPI-CBG)
+49 (0)351 210 1078 (Fax MPI-CBG LMF)
+358 (0) 468102840 (Finnish mobile, only when I'm in Finland)
http://www.bioimagexd.nethttp://www.chalkie.org.uk[hidden email]
(
[hidden email] )