Reconstruction of Overlapped Protein Spots Using RBF Networks

Abstract

Detection and quantification of proteins in two-dimensional electrophoresis gel images is complicated but the same time important task for the development of modern diagnostics and prognostics systems. In this paper the solution of effective protein spot segmentation in two-dimensional electrophoresis gel images is proposed. It is based on the reconstruction of protein spots that are overlapping by the use of modified RBF network. The Anisotropic Gaussian Function with a tilt is proposed for individual protein spot shape model and as the basis function for modified RBF network. By experimentation modified RBF network is proven to be superior to watershed transformation and shape modeling based approaches. The modified RBF network in comparison to watershed transformation separates overlapped spots with the same accuracy in up to two times shorter distance between spots. At the same time computational load of usual and modified RBF networks differs insignificantly. Ill. 6, bibl. 9 (in English; summaries in English, Russian and Lithuanian).