The widespread availability of automated fluorescence microscope systems has led to an explosion in the acquisition of digital images by biologists. This has created a need for computer applications that automate the analysis of these images and an opportunity to develop new approaches to classical problems. An example is the determination of the subcellular location of a protein from immunofluorescence images (or, more recently, images of GFP fluorescence). Current practice is to compare such images to mental images that a cell biologist has developed over time, and to reach a tentative conclusion about the structure (i.e., organelle) that a protein is found in. Since this determination is subjective, it often must be followed up by double labeling with a marker protein from the suspected structure.

As an initial exploration of the feasibility of automating the determination of subcellular location, we developed a system that is able to classify the localization patterns characteristic of five cellular molecules (proteins and DNA) in Chinese Hamster Ovary (CHO) cells. Images were acquired on an epifluorescence microscope after the cells had been fixed, permeabilized, and labeled with appropriate fluorescent reagents (usually antibodies conjugated to fluorescent dyes). The labels used were directed against a Golgi protein, a lysosomal protein, a nuclear protein, a cytoskeletal protein, and DNA.