Image data mining of fruitfly gene expression patterns

Abstract

Understanding the spatio-temporal gene expression patterns is critical in studying genes roles and their complicated relationships. With the availability of high-resolution digital images of in situ mRNA hybridization expression pattern of Drosophila embryos collected through both the Berkeley Drosophila Transcriptional Network Project (BDTNP) and the Berkeley Drosophila Genome Project (BDGP), we will be able to answer a series of interesting questions, e.g., what are the variations of gene expression patterns, what are the patterning process of gene expression, etc. To achieve these goals at a large scale, it is extremely important to automate the image mining and informatics processes of embryogenesis expression patterns. We developed computer programs to register, compare and analyze these spatio-temporal pattern images. In this talk, I will review four projects on image mining of fruitfly embryo gene expression patterns: (a) Gaussian mixture model based embryonic expression pattern extraction and comparison, (b) Gene expression pattern clustering using novel MSTCUT and probabilistic ensemble clustering techniques, (c) Surface/volume models for 3D modeling and registration of early embryos, and (d) A manifold learning method for spatio-temporal registration of 3D gene expression patterns and reconstruction of the embryonic developmental time series.