| Protein subcellular localization is a major indicator of protein function, and efforts have been made to systematically determine the localization of each protein in budding yeast using fluorescent tags. Based on the fluorescence microscopy images, subcellular localization of many proteins can be classified automatically using supervised machine learning approaches. Budding yeast has a stereotypical reproduction mode, such that cell-stage is related to the presence and size of a growing bud. In this work, I investigate the benefits of a cell recognition method and image features that utilize prior biological knowledge of budding yeast shape and its cell-stage dependent changes.;I show that modeling cell-stage dependency of protein abundance and spatial distribution (expression pattern) within a continuous model for cell growth allows the identification of most previously identified localization patterns in a cluster analysis. Further, I show that similarities between the inferred protein expression patterns explain similarities in protein function better than previous manual categorization of subcellular localization. These results suggest that incorporating prior information about yeast morphology in automated image analysis will yield unprecedented power for pattern discovery in high-resolution, high-throughput microscopy images.;Finally, using these new computational methods, I explore cell-to-cell variability in protein abundance and subcellular localization. I define a mean to quantify deviations in subcellular localizations, and find that the method defined is in agreement with previous measurements of cell-to-cell variability in the case of protein abundance. Hence, I show that cell-to-cell 'spatial variability' is a protein expression property, whose measurement is only possible from microscopy images. This measure allows the systematic detection of many classes of such variability, without the use of any prior knowledge about subcellular localization. |
