HMM with auxiliary memory: a new tool for modeling RNA structures

For a long time, proteins have been believed to perform most of the important functions in all cells. However, recent results in genomics have revealed that many RNAs that do not encode proteins play crucial roles in the cell machinery. The so-called ncRNA genes that are transcribed into RNAs but not translated into proteins frequently conserve their secondary structures more than they conserve their primary sequences. Therefore, in order to identify ncRNA genes, we have to take the secondary structure of RNAs into consideration. Traditional approaches that are mainly based on base-composition statistics cannot be used for modeling and identifying such structures and models with more descriptive power are required. In this paper, we introduce the concept of context-sensitive HMMs, which is capable of describing pairwise interactions between distant symbols. It is demonstrated that the proposed model can efficiently model various RNA secondary structures that are frequently observed.