Colocalisation study

[andrea.s...@gmail.com]

Hello,

I need to run some colocalisation analysis with Icy to calculate Pearson and/or Mander’s coefficients. Which one would be the best to present for publications? I am happy to do both

When it comes to Icy, could anyone tell me what is the exact procedure to get the best result and avoid false positive results? I need to process z-stacks of (3 channels) of human cells. 1 of these channels is about cytoplasmic organelles (big in size),the other two are cytoplasmic proteins. 
Below I listed the manual procedure before running the colocalisation studio plugin:
1-Intensity projection
2-Select ROI 
3-Run Colocalisation studio (option correlation) on ROI of channels of interest.

Is there anything missing? It looks too simple to me but it seems to be working. However, in this way I manage to collect only the Pearson coefficient but not  Pvalue as it always results "0", and Mander’s gives me error about the Zstack (don't know why). 

Also, I would like to make this procedure automatic by creating a protocol. 
I do the following:
1-sequence
2-intensity projection
3-extract channel (1 block each)
4-colocalisation studio-correlation (1 block each combination)
5-workbook to file (save in .xls)

This seems to be working, but I would like to apply the protocol to ROIs that I manually drew on original file.
Could anyone help me troubleshooting this procedure, or do you have already a protocol for this I could try to use? 

Thanks in advance and sorry for the long message.

[thibault...@yahoo.fr]

Hi,

I need to run some colocalisation analysis with Icy to calculate Pearson and/or Mander’s coefficients. Which one would be the best to present for publications? I am happy to do both

First of all, there are important issues with Pearson and Manders coefficients: Pearson is sensitive to noise and background intensity and both coefficients are difficult to interpret. You can read this recent review: Lagache t et al, cyometry Part A 2015.

When it comes to Icy, could anyone tell me what is the exact procedure to get the best result and avoid false positive results? I need to process z-stacks of (3 channels) of human cells. 1 of these channels is about cytoplasmic organelles (big in size),the other two are cytoplasmic proteins. 
Below I listed the manual procedure before running the colocalisation studio plugin:
1-Intensity projection
2-Select ROI 
3-Run Colocalisation studio (option correlation) on ROI of channels of interest.

Is there anything missing? It looks too simple to me but it seems to be working. However, in this way I manage to collect only the Pearson coefficient but not  Pvalue as it always results "0", and Mander’s gives me error about the Zstack (don't know why). 

What is the Zstack error? the pValue can be "0" if it is below 10^(-16).. 

If you would like to know if molecules are truly overlapping organelles, I recommend you to use another function that counts the number of spots inside the organelle mask and perform a statistical test to challenge the null hypothesis of molecules random distribution inside the cell. If you are interested, I can make publicy the block I have designed for this?

Thibault Lagache

[impa...@gmail.com]

Hi, 

I am having a very similar problem. Two of my proteins partially overlap when transfected in human cells. I would like to quantify this using co-localisation study. However, it seems that there might be too much noise in my samples since almost every image I use has a Pearson r value of at least 0.5. 

I think I am having the same problem as the query above. How do I avoid getting these sort of false positives?

Currently what I do is. 

- acquire z-stack images with three channels

- (I need co-localisation values for two - the red and green).

- transfer the raw data file (.nd2) into Icy

- use co-localization studio

- select correlation method

- define ROI by polygon and select this region

- define channel 1 as 0 and channel 2 as 1

- run icy 

I subsequently get no p-value, it always seems to be 0.0. Also pearson r values are high for samples which by eye do not co-localise. 

My manders analysis always fails as well.

I tried to send you the raw data files but I get an error. I suspect the files are too large. 

Could you provide any advice on how best to resolve this problem of what I suspect is false positives?

Thanks