In one class of alignment-free approach, relatedness of two sequences is based on the number and value of sub-sequences (e.g. k-mers) they share in common.
To detect putative RNA RNA interactions (for review see [ 24, 25]), a first class of sequence alignment methods, such as BLAST [ 26] or ssearch [ 27], can be used to search for long stretches of complementarity between the mRNA target and the sRNA reversed query sequence.
For the most conserved class of vertebrate alignments, 58% of non coding regions are alignable between human and fish, versus 95% of coding alignments (66.48% identity in alignable non coding sequences, 69.23% in alignable CDS).
In the 2nd class of conserved alignments (dotted lines), the conservation of non coding regions is very similar to CDS in closely related species, and, as for the first class, suddenly decreases in distant vertebrate species.
In this work we focus on the second class of methods, alignment-free algorithms that rely on oligomer frequencies and map two strings X and Y onto corresponding multidimensional vectors X and Y; these vectors are indexed by a number of substrings in the given alphabet (a typical case is when all possible substrings of a predefined length k are used).
The general edit distance algorithm is a classic editing type algorithm for RNA comparison, whereas RNAforester represents the class of tree alignment-based algorithms, which can be due to their working principle much faster, but are less accurate than editing algorithms.
Both classes of MSLL alignments exhibit strongly bimodal distributions, with 5' and 3' peaks and a trough in the gene interior.
But, for the sequence set considered in this study, it worked well: the different types of tree topologies that we observed could be related to distinct classes of sequence alignments.
A previous analysis by Cuomo et al. has shown RIP induced SNPs in this class of transposons however an alignment was not reported.
