Automated Classification of Images of Protein Subcellular Localization for Mass Transfection of cDNAs Clones

N. Kabuyama, R. Minamikawa-Tachino, T. Gotoh, S. Kagei, M. Ishibashi, T. Togashi, S. Sugano, and H. Usami (Japan)


image processing, pattern analysis and recognition, subspace method, bioinformatics


The subcellular localization of protein provides us with useful information about the biological role of protein and its coding gene. cDNAs are also crucial both for annotation of the human genome and for experimental analysis of gene functions. An efficient and accurate procedure for classifying the protein-localization patterns produced by a huge number of cDNAs within cells is required. The protein-localization image was acquired under a fluorescent microscope. This study reports an algorithm to search for protein-localization patterns in the image and to classify the image into standard subcellular compartments. Our algorithm searches for focal points of protein-localization patterns using a model for the patterns, and extracts characteristic features around the focal points, such as gray level value, geometry, and texture. After feature vectors obtained from the patterns were classified using our algorithm based on the subspace method, the image was classified into a subspace with a minimum of mean square errors of all the distances between the feature vectors and each subspace. The effectiveness of our algorithm provided a higher ratio than our target ratio of 90 % for the correct classification ratio in our experiments.

Important Links:

Go Back

Rotating Call For Paper Image