Research On Recognition Of Eukaryotic Gene Acceptor Sites

Recognition of acceptor sites is one of important subjects of bioinformatics. Based onintelligent information processing techniques, this paper researches the recognition ofacceptor sites. The main production of this paper as follows: (1) This paper builds a database of acceptor sites based on HS3D. Then contents of basesequence are analysed and correlation coefficient between coordinate and true result arecalculated. Lastly, this paper analyses statistical result and research results show there is asegment of consensus sequence above AG bases and it can be used to recognize acceptorsites. (2) This paper designs a model of acceptor sites recognition based on LVQ neuralnetwork. The sequence's length is 50 bases and every base is coded by 4 neurons, so thereare 200 neurons in input layer of neural network. The amount of competitive layer's neuronsis 25 and output layer's is 2. Experiment results show that LVQ model can correctlyrecognize more than 70% of the true acceptor sites and more than 85% of the false acceptorsites in the test group. As a method, LVQ model gives some valuable references to acceptorsites recognition. (3) Recognition of acceptor sites has been given attention by many researchers.Research results show that there is relation between acceptor sites recognition and branchsites, but the effect of branch sites on acceptor sites recognition hasn't been specialized.Considering branch sites, this paper specializes in the effect of branch sites on acceptor sitesrecognition based on BP neural network. Experiment results prove that branch sites have animportant effect on acceptor sites recognition. At last, this paper compares BP network withLVQ network based on the same acceptor sites database, result shows that recognitionperformance of BP network excels LVQ network. (4) This paper develops an algorithm of acceptor sites recognition based on motifsequences of acceptor sites. At first, motif database is built by sampling sequences fromlearning group. Then motif sequences of training group are matched with motif database andget some score, the least score of positive samples and the most score of negative samples areregarded as criterion. At last, this algorithm can correctly recognize more than 85% of thetrue acceptor sites and more than 90% of the false acceptor sites in test group. The researches of this paper are partly published on Acta Biophysica Sinica,Proceedings of 2004 World Congress on Intelligent Control and Automation and Proceedingof the 22nd Chinese Control Conference. This paper is supported by CNSF #60234020.