[helix-based] optimize bulge breakpoint

[martin-raden]

I found an example where a shift of a one-sided bulge causes a significant energy change (in --mode=H)

 --> IntaRNA -q GCUGAAAAACAUAACCCAUAAAAUGCUAGCUGUACCAGGAACCACCUCCUUAGCCUGUGUAAUCUCCCUUACACGGGCUUAUUUUUU -t CUGGAUGAAUGACAGGGAAAACAUGCGUAAUACUUACGCAGUUCUCUGAAAAAGUGAUUUAAAUUUAGAUGGAUAGCGGUGUAUGGAAACGUUCUGUUACAUGAAAUGGCCCGUUAGACAUCACAAAUCGCGAAGAGUUUCCCAUUAAUUUUUGAUAUAUUUAAAACUUAGGACUUAUUUGAAUCACAUUUGAGGUGGUUAUGAAAAAAAUUGCAUGUCUUUCAGCACUGGCCGCAGUUCUGGCUUUCACCGCAGGUACUUCCGUAGCUGCGACUUCUACUGUAACUGGCGGUUACGCAC --mode=H

default mode

target
           192                                              226
             |                                              |
5'-CUG...AUUU         A   AAAAAAAUU                  C       ACUG...CAC-3'
             GAGGUGGUU  UG         GCA            UGU UUUCAGC
             |||||||||  ||         |||            ||| |||||||
             CUCCACCAA  AC         CGU            ACA AAAGUCG
3'-UUU...AUUC         GG  CAUGUCGAU   AAAAUACCCAAU   A       -5'
             |                                              |
            48                                              1
query

interaction energy = -14.2394 kcal/mol

helix-based prediction

target
           192                                              226
             |                                              |
5'-CUG...AUUU         A   AAAAAAAUU                  C       ACUG...CAC-3'
             GAGGUGGUU  UG         GCAU            GU UUUCAGC
             |||||||||  ||         ||||            || |||||||
             CUCCACCAA  AC         CGUA            CA AAAGUCG
3'-UUU...AUUC         GG  CAUGUCGAU    AAAUACCCAAUA  A       -5'
             |                                              |
            48                                              1
query

interaction energy = -13.7394 kcal/mol

This can be solved by a special handling of one-sided bulges, for which we only have to consider the addition of non-maximal helices (shifting the last bp to the next helix).

[martin-raden]

thinking about it, this also applies to all other loops... to bypass this problem, we would have to screen all subhelix lengths of the helix to be added. would be ok...

[martin-raden]

ping @RickGelhausen : something we should think about for the journal version of the paper

[martin-raden]

mhh... can only be done easily for the canonical helices... otherwise we have to trace the base pairs of the current helix... not good...

[martin-raden]

the more I think of it, the less good the idea is, since we do only store the energy of the maximal helix and would have to recompute energy terms.

so the alternative is to introduce a "--noLP" into the standard predictor and be done with it.

[martin-raden]

new idea: can be done correctly 👍