Processing 2D Gel Electrophoresis Images for Efficient Gaussian Mixture Modeling

Abstract

In modern molecular biology the most commonly used method to distinct proteins present in complex sample is two-dimensional gel electrophoresis. Unfortunately, the quality of the gel image is reduced by the presence of non-linear background signal, spikes, streaks and other artefacts. The main components of gel image are protein spots. To properly distinguish spots, mostly in overlapping regions, mixture modeling can be performed. Due to many signal impurities the estimation of model parameters is inadequate. In this study, by using two fragments of real gel image and a set of synthetic data, three background correction methods with four image filtering methods were collated and the quality of spot detection based on mixture modeling was checked. The presented results prove that efficient modeling of 2D gel electrophoresis images must be preceded by proper background correction and noise filtering. A two-step Otsu algorithm was the best method for removing background signal. There was no single favorite from filtering methods, but using 2D matched filtering leads to good results despite the background correction method used.