Detection of the Splicing Sites With Kernel Method Approaches Dealing With Nucleotide Doublets

摘要

The RNA-splicing is a very important phenomenon in eukaryotic cells to make pre mRNAs into mature forms. Despite its importance, biochemical details of the mechanism of RNA-splicing have not been understood completely. In this research, using machine learning approaches, we aimed to discriminate the true splice sites from pseudo sites which wouldn't be spliced. Support Vector Machine (SVM) is known as an efficient machine learning method used in many scientific disciplines including bioinformatics [3]. As several researchers have shown, DNA sequence data can be treated with adequate design of kernel functions i.e. core parts of SVM [1, 2, 4]. We focused on patterns of nucleotide doublets of DNA sequence and designed a kernel which might better represent the feature of sequence around splice sites. By examining more than 10,000 human splicing donor (5' ends of introns) and acceptor (3' ends of introns) sites, we found that the accuracies of this methods were 94.49% and 93.94% respectively. These results imply higher efficiency of this method than those observed with other kernel methods and methods based on first- and second-order Markov models. This method would reflect properties of sequence data and would give us insights for the mechanisms of the RNA-splicing.