Hierarchical event in Protein folding

How do you solve this problem in this paper when you want to use sequence to predict structure?

predict structure from sequence

3DPSSM doing pretty good.

algo for hb align

this paper align protein by contact map, which might give us some idea for align protein by hb map.

Predict interlock 8 residues

Maybe we can use Machine Learning methods to predict those 8 residues,
since our theory is those 8 residues will poll the strands together,
nature should be able to recognize it from far away.

beta-turn

since type II beta-turn signal can predict strandon,
it means there do exist pattern in sequences at the turn.
this tells us we also need to analysis the turn, not only the strands.

Type II'. The first residue of these turns has a conformation which can only be adopted by glycine (see below Ramachandran plot). The second residue shows a preference for polar amino acids such as serine and threonine.

predict strandon using beta-turn info

when there's a very strong type II or type II' beta-turn between strand1 and strand2,
then seems like strand1 and strand2 are in the same strandon.

Strandon Prediction

Predict if two consecutive strands are in one strandon or two different strandon.

I have to think a way to predict it.
so I try to gain some idea by reading papers.
1. predict beta turn.
2. predict alpha turn. (use 21D array, PSI-BLAST position matrix) by using similar idea, we might create another position array.
3. predict strands

ideas:
1. maybe we can use HB residue pairs info. tool1 tool2

HB and link bug in 1kul

because PDBSUM info is incorrect, the strands are too short in PDBSUM.
the real strands should be much longer.
this caused very few HB on those strands, and lost link info too.
make our HB alignment very hard.

PAM Blosum

Maybe I'll use PAM or Blosum matrices as my score matrix.

Hydrogen Bond is not 100% Conserved

According to this paper
Hydrogen Bond is not 100% Conserved even within same domain.

motifs which should be deleted.

Motif# 1M, 1N, 1V, 1AB, 2W, 2AF, 2AG, 2AL, 3C, 3J, 4M, 5I, 11B, 13C in SM table should be deleted.
they are not exist in Motif table.

number of structures

should be 749

sql can't use b.1~b.38, because it's text, ex: b.121 will also in this range.

scop is not ordered by it's "b.1.1.1" thing,

the order is based on another file which contents the real order and hierarchy.

we only miss 4 structures. fixed.

problem

there are some structures which we didn't check the strcuture and assume the pdbsum is correct,
actually the pdbsum is not correct, and we need to devide some strands.

ex:
1gof 1-50
1eut 506-647
thus, SM#3R should be deleted.

there should be some other structures have same problems.