Analysis Of Rice Genome Based On Sequence Characteristics And Prediction Of Gene Structure

Computational protein coding gene structures, which is valuable for finding new genes and understanding the composition of genomes, is one of the most important research issues in bioinformatics. However, for rice gene structures, although several gene prediction programs have been developed, the prediction accuracy is still unsatistified.In this study, the features that have been widely used in prediction of gene structures are further analyzed. It's found that the singal profile of functional site signals, the codon usage preference in coding regions and non-coding regions, and the nucleotide compositions in the upstream and downstream regions of splice sites are greatly related to the C+G content of sequences. And then the complicated problem of predicting gene structures in rice DNA sequences is decomposed into a series of sub-problems at several levels with decreasing complexity, including the gene level, the element level (exon, intron, ect.) and the feature level (translation initiation and termination site signals, splicing site signals, codon usage preference, etc.). Further, multilevel models corresponding to different C+G contents are created by a multilevel optimization process, a dynamic programming algorithm is designed to search for optimal gene structures, and a new program RiceGenePRE for the prediction of rice gene structures is developed. Testing results on the widely used dataset OsSNG550 demonstrate that the prediction accuracies of RiceGenePRE at nucleotide level, exon level and gene level are all higher than those of the well-known program FGENESH. A web server of RiceGenePRE is available at (http://bioinfo.hust.edu.cnl).