CP2 versus CP1 IdentifyPrimaryAutomatic coordinates


I am a graduate student doing yeast high-content screening in collaboration with the labs of Brenda Andrews and Charlie Boone. We have been using both CP1 and CP2 to identify images of yeast cells, where our goal is to identify mutants with abnormal DNA damage foci, as well as abnormally formed nuclei.

I have a problem that I need your help with. We have performed a large number of screens using CP version 1 (identifying DNA damage foci), and are now performing screens of those same images using CP2 (identifying abnormal nuclei). I would like to establish a correspondence between the cells identified in our CP1 screens with those in our CP2 screens, to see if there is any interaction between the two phenotypes. Both the CP1 and CP2 pipelines used the IdentifyPrimaryAutomatic module to segment and identify cells. We tried running the attached CP2 pipeline on a test image, and then compared the object.CSV file to one produced from the CP1 pipeline for the same image, but the cell locations did not match. By this I mean the object locations (X,Y) do not seem to match.

Attached are:

  1. 2011_08_22_ver10415_IPANuclei_ISACells_IPACellsStringent_Relate_SaveImages.cp (15.9 KB)
  2. 2011_05_03_ver5811_IPANuclei_ISACells_SaveImages.mat (3.52 KB)
  3. cp1_version_5811_Object.csv (13.9 KB)
  4. cp2_version_10415_Object.csv (20.1 KB)
    What I’d like to do is find out how to map (X,Y) coordinates for CP1 IdentifyPrimaryAutomatic into (X,Y) coordinates for CP2 IdentifyPrimaryAutomatic. Could you suggest how I might do this? I would like to avoid having to re-screen all our data using CP2 if at all possible, as we are trying to wrap this project up.

Thanks in advance for any help you can provide. Please let me know if you need any more information,


Hi Lee,

Essentially, there are three differences you are seeing:

  • The ObjectNumber ordering in CP1 is by ascending X order, whereas in CP2 it is by ascending Y (see the attached plot). So you can sort the CP1 table by Y and produce a new ObjectNumber column that corresponds to it.
  • Correcting for the above, the X,Y locations do not exactly match; they differ by roughly a pixel. Images in CP1 took (1,1) as the origin, whereas in CP2, (0,0) is the origin. So you can subtract 1 from the X,Y locations in the CP1 table to get a closer match (though still not exact to a few decimal places).
  • A few more objects are detected in CP1 than in CP2. This is due to improvements in the thresholding algorithm. There is no CP1-to-CP2 correction/conversion for these objects.