Using SVM to Predict the Protein Spots in 2D-GE Images

Abstract

Proteomics has emerged as a prominent field within bioinformatics in recent years. A proteome is defined as the complete set of proteins expressed by a cell in a given state, and it can vary under different conditions. In proteomic analysis, the initial and essential step is the separation of proteins, for which two-dimensional gel electrophoresis (2D-GE) serves as a crucial and effective technique. With the increasing availability of online image databases, biologists are particularly interested in the information that can be extracted from protein spots. Thus, the rapid and accurate identification of these spots is of great importance. In this work, it proposes a method to separate and detect protein spots in 2D-GE images using the Support Vector Machines (SVMs) technique, with image contrast as the primary feature. First, the image contrasts in 2D-GE images are calculated. Then, the protein spots and background regions are manually labeled within the image. Their contrast values and pixel intensities are subsequently used as training data. A predictive model is then constructed using LIBSVM to identify protein spots in the images. Experimental results demonstrate that the proposed approach achieves promising performance in detecting protein spots from 2D-GE images.