Automatic Analysis Of Gene Expression Patterns Of Fly Embryos

DNA microarrays and RNA in situ hybridization are important method for trackinggene expression. DNA microarray analysis has shown that gene expression analysis hasthe role of understanding of cell function and contact, but ignore the link and complexityof expression patterns, RNA in situ hybridization provide favorable condition forvisualization gene expression patterns.The use of highthroughput RNA in situ hybridization to produce a series ofdifferent stages of gene expression patterns, these patterns with the development of digitalimage technology be recorded in the form of a digital image. Initial in situ hybridizationimage analysis carried out through direct observating microscope picture by the experts,but with the increase of the number of images, analysis of large-scale picture becomesvery complex, so lots of attenions are paid to automated image analysis techniques.Using the BDGP (Berkeley Drosophila Genome Project) public data set as a resourceto build three data sets and to test the word, our research is divided into the followingthree parts:The first part is the image preprocessing of the microscopic images of the Drosophilaembryo in the BDGP database. By digital image processing technology to eliminate thelight and the background in picture, split a large number of embryos in the picture, anpreserve center of independent embryos. Use the longest axis of embryos as a benchmarkto run angle calibration and scale registration,so the last standard image of the idealindependent embryos are obtained.The second part of the work, mainly for the Drosophila embryo gene expressionpattern image clustering. Feature extraction is one of the focus of research in the clusteringproblem. In this paper, the PCA (Principal Component Analysis) and ICA (IndependentComponent Analysis) are combined in image feature extraction algorithm, and aminimum spanning tree-based clustering algorithm is used. Through the experimental,feature extraction and combination of clustering algorithms are better than some otheralgorithms in the application of this subject. The third part of the work, note the fruit fly embryo gene expression picture.Annotation problems due to classification problems in pattern recognition, and selectedfeature selection method of related for the category. By two-dimensional DWT (DiscreteWavelet Transform) processing, extract the image of the wavelet transform coefficients asthe characteristics of the image, and then select the optimal characteristics for annotationswith methods based on Relief. Each image in BDGP data setcorresponds to a range ofannotation vocabulary, this multi-objective problem is solved by a series of parallelclassification classifier.