One task I routinely have to do is evaluate the effect of variants found in the UTRs of human mRNAs. One site I use for this is RegRNA 2.0 in the hopes of finding transcription factors or new miRNA sites (in the case of 3'-UTRs). Something I have recently been tasked with is evaluating if a change affects RNA folding. While RegRNA has an "RNA accessibility" measure, which I am not sure how to interpret, I have become more familiar with two of the packages at Vienna's RNA server, RNAfold and RNAalifold. I am trying to determine the right way to use these two packages. Currently, I will take ~100 bases or so around the site, blastn them, and pull orthologs (including predicted aka XM) that have more then 90% identity. I use clustw to align them, remove sequences that have indels larger then 4 base pairs, and then finally input this alignment into RNAalifold. Then I will use the constraints from this prediction to fold the wild type and mutant sequences with RNAfold. Of the handful of times I have done this I have seen no changes.
How big should I make my blast query? My assumption is that local structures would be accurate, given a large enough input. Here is the only similar question I found on Biostars. From my experience, I have found the folding rpediction changes when I change the size of the input. Furthermore, are there appropriate settings in RNAfold/RNAalifold that need to be changed to answer this kind of question? For example I notice a lot of 5' and 3' pairing, which wouldn't necessarily be accurate since it is a snippet of a larger RNA.
Please let me know if this question violates guidelines or rules and I will gladly remove or reformat it.
It appears that I finally hit the right search term. RNAplfold might be the better package for this kind of question.