A new algorithm for predicting splice site sequence based on an improvement of categorical discriminant analysis.

The splicing signals which govern the excision of introns are not yet well explained, since actual splice site sequences are to some extent different from the generally accepted consensus sequences. While a quantification method (categorical discriminant analysis: CDA) has been proposed to analyze splice site signals, sample sequences lying in the overlapping region of sample scores are not discriminated well by CDA, thus limiting the predictive ability of this analytical technique. In this paper, we propose a method to improve the performance of CDA, which is applicable to the analysis of 5'-splice site signals in various mammalian genes. The proposed algorithm has revealed a distinct enhancement in the predicting ability compared to that of traditional CDA and could explain point mutations in the 5'-splice site region of rabbit beta-globin gene fairly well.