Fluorescence Imaging for Life Sciences forum: topic
This is a public forum
Measuring Colocalization
Paula Nunes
Friday, 22 February 2008 13:52 UTC
Sounds like the most basic thing a microscopist should know how to do. Ok. Well, to be honest I am really confused!
We have the software package Volocity (Improvision) which has a very fancy interface that lets you choose a threshold in each channel and then it spits out like 7 different types of correlation coefficients…
the problem is that depending on what you choose as your threshold you can get completely different numbers!
Can anyone comment on what the best way to choose a threshold is?
Another question of interest is about pre-processing. Of course everybody wants to quantify the raw data, but it seems to me that noisy pictures will give you false colocalization because you are measuring the colocalization of the noise too. Is it really faux pas to denoise your images before measuring? What do other people out there do?
-
Replies
-
See Bolte & Cordlieres, J. of Microscopy 2006 224:213-232 for an excellent review on how to do co-localization.
Best, Jennifer
-
Personally, I would never de-noise or “smooth” the data prior to calculating co-localization. This can be very dangerous and misleading. In the confocal setups that you mentioned from the other forum topic, there should be minimal co-localization of your noise from different channels. The “dark counts” of your PMTs should not really overlap due to their random nature, and would only provide a negligible contribution to your co-localization quantification. In turn, if you are comparing a couple different markers on the same microscope set-up with the same acquisition parameters, the noise overlap will provide a similar “background” in all of your co-localization counts, so it shouldn’t be a problem, especially if you employ a background subtraction that I will briefly describe below.
If you see significant co-localization in the “noise”, it is possible that you are actually looking at an artifactual signal that you should isolate and eliminate from your sample or microscope set-up.
For proper quantification of confocal images (or any collected data, for that matter), it is important that your signal not be saturated at the “high” end of the spectrum, and the threshold should be basically a background subtraction of your sample. Within your images, you could select a region of interest (ROI) outside of the field of your sample, calculate the average pixel value within the ROI, and subtract this number from all of your pixels. I’m sure that your software has some kind of function that would be similar to this, but since I am not familiar with it, I can’t say for sure.
Good luck!
-
I agree with Noah. But be careful not to confuse background (which can be subtracted) with noise. Noise causes a variance of the pixel intensity value around the “real” value. So the only good way to “denoise” your images is to use frame averaging.
-
Excellent point, Jennifer, thanks for the clarification.
-