Protein Spot Matching Algorithm Based On Protein Area SIFT Feature

The Comparative analysis and research of protein spots in2D gel images is animportant research contents in the proteomic research. Protein spots matching in2D gelimages is the key technology in comparative analysis and research of protein spots.Since the traditional matching algorithm can hardly work when matching rotary andskewed gel images and are weak universality, this paper studied the protein spotsmatching method deeply in2D gel images, sponsoring by the project of Natural ScienceFoundation of China and Natural Science Foundation of JiangXi Province, The researchwork and achievements are as follows:1. Summarize the basic theoretical knowledge which commonly used in proteinspots matching in2D gel images. Firstly, common matching features are analyzed.Basing on the characteristic of gel images and the distribution of protein spots in gelimages, this paper use graph features and SIFT feature as major features. Secondly,similarity measures and how to choose metrics formula basing on features that used inthis paper are introduced. Then, common space transformation models and how toachieve the model used in this paper are introduced. Finally, how to choose andoptimize search space and search strategy are analyzed.2. A protein spots matching algorithm based on graph features was presented.Firstly, GG and RNG was created based on the info of the coordinate of protein spots.Secondly, robust variance features was derived by statistic method that analyzed thedegree of vertex and the distance between vertexes. Then, global matching wasprocessed based on variances features in gel images to find root node. Finally, basing onmaximum relation spanning tree used as comparing architecture and gaussian-typesimilarity measure used as metrics formula, local matching was achieved. Experimentalresults which based on some source images showed that the proposed matchingalgorithm had well matching effect for gel images with high quality, but had badmatching effect for gel images with poor quality, the proposed algorithm need to make afurther study.3. A matching algorithm based on protein area SIFT feature was proposed. Firstly,SIFT features were extracted in protein area. Then protein spots were matched coarsely by use of SIFT features in protein spots area and mismatched features were removed bymeans of RANSAC. Finally, the correlation of TPS transformation was obtained basedon the result of coarse matching. The precise matching in the rest of protein spots whichis not matched in the process of coarse matching, is accomplished by the method ofgeometric correlation. The algorithm is applied to some source images including theinternational gel images and the gel images of Bio-Rad Company. Experimental resultsshow that the proposed matching algorithm has higher matching accuracy, the matchingerror is less than2.2%, and as well as better stability especially for rotary and skewedgel images.