ref papers

this paper use some probes about 5~9 residues long to predict the structures.
it's like in the middle of homology modeling(lots of residues) and my previous finding which use 4 residues to predict loops.
they obtained the probes by statistics similar to HMM.

this paper use similar approach as mine, they try to predict Strand-Loop-Strand structure.
I think they got a pretty good result, and I might use their result as input.

this paper try to predict folding initiation sites. which might solve our problem by this way.

this paper predict long range contact by contact maps.
I think this might help a bit in our case.