Gap treatment
In the new variant of
RNAalifold, gaps are not used for energy
evaluations. That is before e.g. the energy of a hairpin loop is
determined, all gaps are removed and the length of the hairpin used to
compute the energy contribution is the length of the hairpin in the
respective sequence.
Not using gaps in energy evaluations is especially useful if a vast
majority of sequences of an alignment share these gaps, and only some
have insertions that will lead to interior loops. An example of how the
new treatment of gaps increases the predictive power
of
RNAalifold can be seen here:
-
Prediction of GcvB structure
Covariance evaluation using RIBOSUM derived scores
Instead of using a hamming distance based score, we developed a scoring
based on RIBOSUM matrices. There are two major atvantages of using these
scores:
-
Scoring matrices can be chosen to fit to the alignment in
question.
In alignments with different pairwise identities the
probabilitites to get consistent and compensatory mutations vary. The
scoring can now reflect that.
-
In contrast to a hamming distance based score, having no mutations can
also lead to a slight bonus.
Examples of how RIBOSUM based scores improve the predictive power
of
RNAalifold can be seen here:
A combination of both effects can be seen in: