RNA secondary structural alignment with conditional random fields


MOTIVATION: The computational identification of non-coding RNA regions on the genome is currently receiving much attention. However, it is essentially harder than gene-finding problems for protein-coding regions because non-coding RNA sequences do not have strong statistical signals. Since comparative sequence analysis is effective for non-coding RNA detection, efficient computational methods are expected for structural alignment of RNA sequences. Several methods have been proposed to accomplish the structural alignment tasks for RNA sequences, and we found that one of the most important points is to estimate an accurate score matrix for calculating structural alignments. RESULTS: We propose a novel approach for RNA structural alignment based on conditional random fields (CRFs). Our approach has some specific features compared with previous methods in the sense that the parameters for structural alignment are estimated such that the model can most probably discriminate between correct alignments and incorrect alignments, and has the generalization ability so that a satisfiable score matrix can be obtained even with a small number of sample data without overfitting. Experimental results clearly show that the parameter estimation with CRFs can outperform all the other existing methods for structural alignments of RNA sequences. Furthermore, structural alignment search based on CRFs is more accurate for predicting non-coding RNA regions than the other scoring methods. These experimental results strongly support our discriminative method employing CRFs to estimate the score matrix parameters. AVAILABILITY: The program which is implemented in C++ is available at http://phmmts.dna.bio.keio.ac.jp/ under the GNU public license.

Bioinformatics, 21(Suppl 2):ii237–ii242