Structure-based realignment of non-coding RNAs in multiple whole genome alignments

Abstract

Whole genome alignments have become a central tool in biological sequence analysis. A major application is the de novo prediction of non-coding RNAs (ncRNAs) from structural conservation visible in the alignment. However, current methods for constructing genome alignments do so by explicitly optimizing for sequence similarity but not structural similarity. Therefore, de novo prediction of ncRNAs with high structural but low sequence conservation is intrinsically challenging in a genome alignment because the conservation signal is typically hidden. This study addresses this problem with a method for genome-wide realignment of potential ncRNAs according to structural similarity. Doing so reveals thousands of new high-confidence ncRNA predictions with particularly low sequence conservation from an alignment of 12 Drosophila genomes and hundreds from an alignment of 28 vertebrate genomes in the Encode project.