A Novel Approximate Inference Approach to Automated Classification of Protein Subcellular Location Patterns in Multi-Cell Images

Abstract

The subcellular location of proteins is most often determined by visual interpretation of fluorescence microscope images. In recent years, automated systems have been developed so that the protein pattern in a single cell can be objectively and reproducibly assigned to a location category. While these systems perform very well at recognizing all major subcellular structures, some similar patterns are not perfectly distinguished. Our goal here was to improve performance by considering more than one cell in a field. We describe how to construct a graphical model representation for a field of cells while taking into account the characteristics of the cell type being studied. We show that this approach provides improved performance on synthetic multi-cell images in which the true class of each cell is known, and that a new approximate inference method can provide this improved performance with significantly faster computation times than previous approaches.