We were not pleased with the alignments that came out of the
fold-recognition methods, since most of them ended up having predicted
helices in the strand positions.  Although it is quite possible that
our secondary structure predictors were consistently wrong, we felt it
more likely that the alignments were thrown off by insertions---a
common problem with TIM barrels.

We tried using constraints on the sheets in undertaker to impose the
strand pair that we believed in.  This did not work very well, since
we never got a completely consistent model to form the sheets we
wanted.


This is try12-opt2, which reflects a different belief about where on
the sequence the 4th strand is.