Idendifying substructures of actin cytoskeleton

cellprofiler

#1

Hey there,
my name is Matthias Truttmann.I am a PhD student in the Dehio group at the Biozentrum, Basel (Switzerland). Currently, I am setting up a high-troughput screen using primary cells (or Hela) to study host-patogen interactions. I am using CP since half a year and the program works in general great!
However, I have encountered a problem while establishing a pipeline suitable for analysation of results comming from a small test screen.

My target structures I’d like to have idendified (invasomes) appear in different shapes and different sizes. They can look like a bowle / circle or appear ring-like. Unfortunately, there is no way to stain them specifically as they represent just special actin rearrangements.

If appling a Threshold, I get up to 90% correct idendifications for a test image set. However, running the pipeline on a real image set, I get a false positive rate of more than 100% due to the fact that huge compartiments of the “normal” Actin skeleton pass the threshold filter and finally look kind of similar in shape and size to my target structure - at least for CP. I posted a few example images under the links below. images A02 contain the target structure, while all other images do not.
Using primary cells, the false positive rate is even higher! Thus, my pipeline does not work satisfying yet. (The pipeline is also posted as .txt file)

How would you try to idendify and count this ring-like structures?

Another problem: CP often splits one single invasome structure into several compartiments, which get finally counted as individual objects. As my target structures are not equal in size, I cannot solve that problem by increasing the minimal allowed diameter in the IdendifyPrimary module. In my specific case, neighbouring objects which are really adjacent to each other (less than 3 pixels in between both objects) do not exist. Thus, I’d like to combine all objects with their direct neighbours that get idendified using the MeasureObjectsNeighbours module. This would dramatically decrease the false positive rate.

Is it possible to do that using one of the currently available CP modules?

I hope you can give my some help! I would really like to use CP for the analysis of the screen as it allows my to extract tonns of parameters in no time.

greetz
Matthias

yourfilehost.com/media.php?c … _s6_w2.jpg
yourfilehost.com/media.php?c … _s3_w2.jpg
yourfilehost.com/media.php?c … _s7_w2.jpg
yourfilehost.com/media.php?c … _s9_w2.jpg
yourfilehost.com/media.php?c … _s1_w2.jpg
yourfilehost.com/media.php?c … _s1_w2.jpg

yourfilehost.com/media.php?c … peline.txt


#2

Hello,
Sorry for the long delay! We have recently expanded our team greatly and have been training everyone and neglecting the Forum. Have you solved this problem, or is it still as described?

Anne


#3

Hey Anne,
in the meantime, I got used to CP and Mathlab programming, thus I got much closer to solve my specific problem. Ben Misselwitz (ETH, Zürich) is giving me a helping hand if I am in troubles.

thanx anyway
Matthias


#4

Hello Matthias,
Feel free to let us know if you need any help in the future.

~martha