Automated classification of subcellular localization of proteins in HeLa cells using fluorescent microscopic images

Abstract

We have developed a system to classify cellular forms of fusion proteins with an enhanced yellow fluorescent protein (EYFP) into subcellular compartments using images by a fluorescent microscope. The system aims at image classification to accommodate the multiplicity of cellular forms. The cellular forms automatically searched for in the images are classified with excellent reliability using statistical pattern recognition. The patterns are extended models of the cellular forms which are the result of protein localization as seen in seven standard subcellular compartments, including in the apoptosis and the overexpression. The image itself is classified by the majority of cellular forms it contains so as to reflect the characteristics of cell population, rejecting cellular forms with low reliability and in the apoptosis. We have found that our system is 97.9% accurate in classifying cellular forms into subcellular compartments. Copyright 2004 Chem-Bio Informatics Society.