We proposed a novel method to deliver kernels for RNA sequence data using stochastic context free grammar (SCFG)¹⁾. Our previous work was to deliver kernels for general biological sequences using hidden Markov model (HMM)²⁾. RNA sequences can not be dealt with HMM because they involve remote base interactions which consequently form stem-loop structures. The stem-loop structure thermally stabilizes secondary structures of RNA, which is essential in terms of evolutionary conservation. SCFG is more powerful stochastic language model than HMM which allows dealing with the stem-loop structures (Fig1). We call our novel kernel Marginalized Kernel over SCFG. The kernel shows good performances in several demonstrations. Fig2 shows a result of kernel PCA for three-class human tRNAs.



Related Information
1) T. Kin, K. Tsuda and K. Asai: "Marginalized Kernels for RNA Sequence Data Analysis", to appear in Genome Informatics 13, 112-122 (2002)
2) K. Tsuda, T. Kin and K. Asai: "Marginalized Kernels for Biological Sequences", Bioinformatics, Vol. 18, Suppl. 1, S268-275 (2002)

Tsuda, K., T. Kin and K. Asai: "Marginalized Kernels for Biological Sequences", Bioinformatics, Vol. 18,Suppl. 1, S268--S275(ISMB2002), 2002.

Kin, T., K. Tsuda and K. Asai,"Marginalized Kernels for RNA Sequence Data Analysis", to appear in Genome Informatics 2002.