Clustering of conserved RNA secondary structures. Candidate human ncRNA sequences were obtained from the RNAz screen of Washietl et al. (2005). Twelve hundred ncRNA candidates were obtained by removing overlapping sequences and sequences with inconsistent genome coordinates, and retaining a single sequence per paralogous set. Sequences were folded using RNAalifold (Hofacker et al. 2001) and then compared pairwise using RNAforester (Höchsmann et al. 2004) using normalized scores for each comparison. RNAforester options were local comparison and no use of sequence information (no Ribosum). The resulting 1200×1200/2 matrix was then used to create an UPGMA cluster using mean values. Tick marks at bottom indicate position of known RNA genes and of non-miRNA hairpin structures (diffuse hairpins).
