Human Promoter Recognition Algorithm

A challenging task is to find the genes and their regulatory network after the human genome draft was completed. Promoters are important elements for gene expression and regulatory and play the important rule in recognizing genes. Human promoter recognition technology today has become a hot spot. And it has very important theoretical significance and practical value.Based on large numbers of publications, this thesis studies human promoter recognition technology and proposes two novel human promoter recognition algorithms.A novel human promoter recognition algorithm based on KL divergence and BP neural network is proposed. First, the most effective 6-mers are extracted by KL divergence, and their frequencies are chose as the component features. Then CpG islands features are computed. The CpG islands features and the component features are combined and form the feature vector that distinguishes promoter sequence regions from other DNA sequences regions. Finally, BP neural network technology is used to construct a human promoter classifier. The classifier consists of Promoter-Exon sub-classifier, Promoter-Intron sub-classifier and Promoter-3'-UTR sub-classifier. Each sub-classifier is a BP neural network. The results of the three sub-classifiers are integrated to recognize a promoter sequence.A novel human promoter recognition algorithm based on two-level SVM classifier is proposed. Support vector machine technology is applied to design a two-level SVM classifier. The first level SVM classifier judges DNA sequences by CpG islands features. If a DNA sequence is not regarded as a promoter by the first level SVM classifier, then it will be recognized again by the second level SVM classifier. The second level SVM classifier consists of three sub-classifiers, namely Promoter-Exon SVM sub-classifier, Promoter-Intron SVM sub-classifier and Promoter-3'-UTR SVM sub-classifier. Each sub-classifier recognizes promoters by the component features. The results of the three SVM sub-classifiers are integrated to predict a promoter sequence. All promoters recognized by the two-level SVM classifier are final experiment results.The experiment results show that the proposed algorithms are effective with high sensitivity and specificity.