Prediction of transcriptional start sites of genes using asymptotic local approach

A popular approach to detect the Transcriptional Start Sites (TSSs) of genes that are CpG sensitive is based on the CpG islands. For these genes, their TSSs is characterized by sudden increases in the CpGs in the DNA sequence. In this paper, a novel gene prediction method is proposed that transforms the problem of detecting the TSSs to that of detecting a change in the mean of a stochastic process using the asymptotic local approach. Features of the CpG islands, such as the cyclic nature, are used to reduce the false detection rate. The proposed method is applied successfully to identify all the genes in the rabbit alpha-like gene cluster, and in a section of the human chromosome 22, 73% of the confirmed genes are predicted. Comparison with the Dragon Gene Start Finder is also made showing that the proposed method has a higher sensitivity.