Colonies intensity (GFP)

Hi everyone!
Congrats for this forum, specially for the whole community around cell profiler.
First, I’m more a sequencing guy, so sorry for my naive perspective about image matters and comments/questions.
Here is my post: I’m trying to differentiate colonies (fungi) expressing GFP from the ones which don’t (mutants) on a petri dish. In theory, the problem seems easy to solve, though I’m getting steps-back on the way. We’re talking about look at hundreds of dishes and thousands of colonies; so I’m looking into a more computer based solution. And my second biggest problem is that non-fluorescence colonies have a rather auto-fluorescence which make it harder (not to mention the auto-fluorescence coming from the media).
So, my pipeline at first was to take one pic with epifluorescence light (using BioRad Chemidoc MP) of the plate. Then use the same plate with white light and with the help of ImageJ (after background correction and transform into grey scale) using an Image Ratio plugin to determine which of the colonies present (whites) habe more (GFP expression) or less (mutants) intensity on the epifluorescence image. So far, not really good results.

Then I met Cell Profiler, and the all its capabilities. I have checked a couple of protocols, like Yeast colonies measurement cp project or colocalization cp project. But I’m not sure if I’m going in the right direction. Any input would be appreciated. I can attach a couple of pics, so you can see the nature of the problem.

Colonies-GFP after background correction and grey conversion


tip: the ones on the left side seem to be more intense :wink:

Colonies-White after background correction and grey conversion

Thanks

Hi,

Your plan sounds fine – put even more simply:
(1) Use the white/brightfield image to segment/identify all the colonies
(2) Measure the GFP intensity of each of the colonies from #1.
Right?

When I look at your white and gfp images, however, it seems like all colonies express GFP. This seems a little odd, though,given your premise, or is it simply the case for this example?

Also, what is the large oval-shaped artifact (reflection?) on the gfp_grey image? It may cause image analysis issues.

The circular plate edge will cause issues and this is usually best to mask out. This can probably best be done using a non-background-corrected original image, which can often help to define the inside versus outside of the plate. Use an ApplyThreshold to get a binary mask for the whole plate that you can use later to mask out the plate and regions external to it.

Feel free to post your pipeline attempt, plus some original images, too.

Thanks,
David

Thanks for the input David!
Yes, that’s a good plan. Though as you said, all the colonies look like they express GFP, but they don’t. On the GPF image, the ones on the left have the contruct and GFP is induced, while the ones on the right size are wild type (no GFP, nor expressed, nor even exist in the genome). So that’s the big challenge, differentiate both types based on intensity, which difference is not striking but it exits.

Sorry about the oval-shaped artifact, it’s produced by the BioRad Chemidoc lights over the edge. It can be solved with ticker media.
Attached is an original GFP image. Thanks


Hi @frodriguez,
Just following up on this and it wasn’t clear to me re-reading this – did you want more help? Thanks!

Hi David,
Thanks for the follow-up. I kinda still working on it. It’s like a side of a side project; and as I said, Image processing is not my thing. Cell profiler is a tool with immense potential, but I have to spend more time on it.
On my way to solve the problem, GE offered a demo of one of their machines: AI600RGB, which is able to detect fluorescence at colony level. It also came with a software (ImageQuant or something like that) to calculate intensities per detected colony which has help me me a lot to deal with the initial step of the project. I’m not trying to promote their product.
Once again, thanks for your inputs. Hope one day I can code pipes for images as I do for sequencing.
Best
Fernando