Developing a quantitative, cellular resolution morphology and gene expression atlas for Drosophila embryogenesis: towards a digital 'Campos-Ortega and Hartenstein. Soile V E Keränen¹, Jonathan T Barron², Pablo Arbelaez², Jitendra Malik², Mark D Biggin¹, David W Knowles¹. 1) Life Sci Div, Lawrence Berkeley Natl Lab, Berkeley, CA; 2) Electrical Engineering and Computer Science, UC Berkeley, Berkeley, CA.

   Animals comprise dynamic 3D arrays of cells, differing in histological type, shape, size, location, etc. We are extending our VirtualEmbryo (http://bdtnp.lbl.gov/Fly-Net/) of the cellular blastoderm to create a quantitative, digital, cellular resolution atlas of morphology and gene expression for all of Drosophila embryogenesis. Because late-stage embryos have some 40,000 cells, 70 cell types and major organs, we have had to develop new strategies to analyze these complex morphologies. Using stage 16 embryos as a model, we have developed an analysis pipeline to automatically find whole tissues or organs in embryos stained to label nuclei. First, target tissues are hand annotated in images of embryos on nuclear stain channel. Annotation accuracy is confirmed by comparison with tissue specific mRNA stains. The annotations are then converted into 3D tissues models, which are used to train support vector machine analysis that automatically locate the tissues in subsequent test images, i.e. images that were not used for model training. We have currently built classifiers that accurately detect six different tissue types based on embryo morphology. Our current detection accuracy ranges between 57% and 87% compared to hand annotations. For example, in the pharyngeal muscles the classifier finds ~80% of the estimated 8914 cells. Repeating such analyses for all tissues will allow us to create an average morphological map and quantitate the differences between individual embryos. Our computational embryology goal is to create maps of each stage of embryogenesis, annotating all tissues, and to link these via computational analysis of fixed and live cell images.