Figure 2
From: Robust and accurate prediction of noncoding RNAs from aligned sequences
Example of the Type B errors and its influence on the prediction methods. (a) Native alignment consisting only of ncRNAs. An alignment of 10 miRNA precursors is highly conserved in terms of both primary sequences and secondary structures. The consensus secondary structure predicted by RNAalifold [33] exhibits a well-known hairpin loop. Profile BPLA kernel and the other prediction methods accepted this alignment. (b) Alignment with the Type B errors. Three miRNA precursors in the native alignment were replaced with unrelated sequences, which destroyed the consensus secondary structure. This alignment was rejected by RNAz due to the drastic decrease in the SCI and also missed by Profile stem kernel. Profile LA kernel was completely ruined showing the higher SVM class probability for the erroneous alignment than that for the native one. Profile BPLA kernel was the only method to accept the alignment by the moderate decrease in the SVM class probability from the native one. Note that the mean pairwise identity is still high allowing this alignment to be produced by sequence-based alignment tools.