Our fold-recognition was pretty useless for this protein, picking up
only a single helix or single hairpin.

We ended up constructing this one mostly by adding sheet constraints
for our predicted strands and helix constraints for the predicted helices.
With only 5 strands and 2 helices, the number of possilbe topologies
seemed limited, but we couldn't get any of the obvious ones to click
into place easily.

We tried using contact prediction to aid in the choice of toplogy, but
ended up with several adequate and no really good models.

This is model try18-opt2.  Topology is s2^vs1||s5^vs3^vs4.
Strand5 is broken but otherwise this looks good.