The challenge of the post-genomic era is functional genomics, i.e., understanding how the genome is expressed to produce myriad cell phenotypes within the context of systems biology. To use genomic information to understand the biology of complex organisms, one must understand the dynamics of phenotype generation and maintenance. A phenotype is the result of selective expression of the genome. It is an expression of the history of the cell and its response to the extracellular environment. In order to define cell "phenomes", one would track the kinetics and quantities of multiple constituent proteins, their cellular context and morphological features in large populations. Such studies should also include responses to stimuli so that functional models can be generated and tested. The key requirement is to define a taxonomy of protein localization in reference to organelle of interest, develop computational methods to quantify their response, and design a multivariate representation of cellular response in context that enables higher level analysis and profiling. The main contributions of this dissertation thesis include: 1) Proposed algorithms for the segmentation of nuclear membrane protein. Based on the observation on the images, we extended the existing iterative radial voting algorithm to iterative tangential voting algorithm for the detection of nuclear membrane signal. After that, some geometric constraints combined with the nuclear information were leveraged for the segmentation of nuclear membrane. This approach enables the analysis of membrane protein on a cell-by-cell basis and was demonstrated to be robust by our experiments. Furthermore, the analysis results have reasonable biophysical interpretation. 2) Proposed a segmentation algorithm for the cell culture on a single focal plane. This algorithm is a combination of geometric constrains and active contour model. It has the following three steps: detection-constraint-evolution, where the detection is performed by iterative radial voting. It enables the analysis of cell culture model on a single focal plane and was evaluated by our experiments. 3) Proposed a segmentation algorithm for 3D cell culture model. This algorithm is the combination of geometric constraints and Radon transform. It has three sub-steps: detection-constraint-evolution, where the detection is performed by iterative radial voting. Since 3D cell culture model has similar functional properties as those observed in human, it provides more reliable information than the analysis based on 2D cell culture model. The robustness was demonstrated by our experiments. 4) Proposed a multivariate feature representation, and gave an example of analysis of membrane protein based on this representation. This representation enables the higher-level interpretation of the bioinformation and the experiments had valuable biophysical meanings. All of our features for different type of data are organized by this representation.
修改评论