Automated Image analysis of Multiple Gene activity patterns in developing animals. 

William Beaver, David Kosman, Gary Tedeschi, Adam Pare, Ethan Bier, William McGinnis, Yoav Freund. 
1) Department of Computer Science and Engineering, UCSD, La Jolla, CA 92093 
2) Cell and Developmental Biology Department, UCSD, La Jolla, CA 92093

Combinatorial Transcriptional Fluorescent In Situ Hybridization (CT-FISH) is a confocal 
fluorescence imaging technique enabling the detection of multiple active transcription units 
in individual interphase diploid nuclei. Current CT-FISH methods in our laboratories are able 
to measure the activity levels of 5 genes in a single embryo or tissue section. Improved 
combinatorial labeling methods will allow simultaneous measurement of twenty gene activities. 

Transforming confocal image stacks with multiple gene activity patterns into usable data 
is a labor intensive task that calls for computational analysis. Our analysis involves: 
1) Segmentation of the cell nuclei using a variety of fluorescent nuclear markers, 
2) Detection of transcription sites on chromosomes, removal of false positives, and 
   classification of nascent transcription sites of specific genes by their fluorescent 
   combinatorial codes, 
3) Registration of image stacks from different embryos to a developmental model, in the 
   process assessing the consistency of the temporal and spatial gene activity patterns. 

Our image analysis involves a combination of image processing and machine learning algorithms. 
The machine learning algorithms allow experimentalists and computer scientists to reiteratively 
tune and improve the analysis system to reflect biological reality. Using such algorithms, we 
show that experimentalists can now overcome the initial analysis bottlenecks, automating 
steps 1 and 2, and in the near future we will implement step 3 and to develop detailed and 
accurate expression atlases for Drosophila embryos and other complex tissues.