I am trying to count x gal positive cells using cell profiler using the counting yeast colony template. Is the any kind soul, who can give me some technical help

Image Analysis
#1

Sample image and/or code

Background

There are blue, brown dots here. I want to count blue dots, those are positive for x gal. These
are reporter T cell, can produce beta-galactosidase.

Challenges

ExampleYeastColonies.cpproj (617.2 KB)
I have tired to customise the below pipeline as above.
ExampleYeastColonies.cppipe (20.3 KB)

I just need to know how many blue dots are there… I will really appreciate if anyone of you can give me some feedback.

#2

Can you explain a bit more what issues you are having that you need help with? Is it finding the colonies (and if so, is it all colonies or a subset that aren’t being correctly identified), measuring the color, etc? It will be easier for someone to assist you if they have a better sense of what issues you are facing, what aspects are working well and which need more optimization.

1 Like
#3

Thank you Beth. In my case, I want to find number of blue dots. And for some reasons, the pipeline is giving weird numbers. Honestly, I am pretty naïve in image analysis, I do not know if I have transferred my message properly.

#4

Sure, let’s break this down a little bit. The beginning of troubleshooting is always just agreeing on what specifically you’re actually trying to see.

  • The template you’re working from is trying to count all spots, then figure out which ones are white vs red; is your goal EXACTLY the same except white vs blue, or do you only want to count the blue spots and don’t care about white at all?
  • What do you mean by “weird numbers”? Too high, too low, etc? Are you comparing them to a hand-count, or is it just your impression that the number is wrong?
  • When you look at the outputs of your object processing modules (such as IdentifyPrimaryObjects, FilterObjects, etc), does it look like you agree with what it is doing, or not? If not, what specific errors are you seeing (aka “in lots of areas, one spot is being called as many spots”, “at the edge, some spots are being missed altogether”, “I agree with everything it’s doing in Image1 but in Image2 it calls too many spots”, etc)? Looking at what the program is doing is ALWAYS the most critical aspect of image analysis.
2 Likes
#5

Thank you very much for your reply. Actually, I want to count only the blue dots. Regarding the output of primary object identification, it seems, In my wild type condition (schizont 1, it is counting less than the treated condition, schizont.2dg1. I am attaching the image and output files for your kind consideration.

Shafi 1
Shafi 11366×622 115 KB
Shafi2
Shafi21366×622 115 KB

I am afraid, I am doing some mistakes regarding typical diameters of objects. I played with that a bit, In the attached image, it is 1 and 8. What do you think I am missing here? Thanks a lot in advance.

#6

It doesn’t look like you changed much from the example pipeline for what you uploaded; I think it is worth you perhaps looking at the included PDF and/or the module notes if you want to get a better sense of what each individual module was supposed to be doing, and therefore where yours went wrong. The image in your screenshot that you showed doesn’t look like what it ideally should before going into IdentifyPrimaryObjects (a pretty dark background with your colony spots looking bright), so it’s not surprising that it isn’t necessarily doing a good job.

A simple pipeline on one of your images that worked well for me, without even needing a template, was this-

  1. ColorToGrey, to split out your channels
  2. ImageMath using the Invert option to invert your Blue channel, making the blue spots now bright-on-dark
  3. EnhanceOrSuppressFeatures with the Speckles option to bring out the spots and delete almost all of the background
  4. IdentifyPrimaryObjects to call the spots.

(Plus then you’ll want probably want some SaveImages/Export/etc modules to write out your results.

The exact tuning should be up to you, because I don’t have a good sense which of the borderline spots are real or not, but it looked pretty decent to me in the View Workspace mode. I can’t promise it will work for all images; you may want to use a plate template or something ala the original pipeline, but that should be enough to get you started.

Good luck!

image
image934×664 226 KB

1 Like
#7

Thank you very much Beth. I really appreciate your response. I was trying to follow your instructions. Your instructions seem very nice and perhaps will work for all. May I ask a bit more about the primary object identification setting you used? I have been trying for last several hours, but in vain. Thanks a lot. I could not appreciate it more.

Regards
Shafi

#8

I didn’t save the pipeline, but I’m reasonably certain I used RobustBackground thresholding and a smallish foci size, like 3-20 maybe?

It should be obvious from the outlines though whether or not it’s circling too much stuff or too little (in which case, you need to adjust your threshold), and whether or the circles are pink or green (and therefore whether your sizes are right). If you’re having trouble understanding how that module works, you may find this video helpful.