We made conjectures about the disulfide bonds, and later confirmed
them from the literature.  We put in 8 disulfide constraints in all
our models (except the initial fully automatic one):

SSBond	C150	C120
SSBond	C178	C166
SSBond	C205	C52
SSBond	C312	C240
SSBond	C321	C223
SSbond	C346	C402
SSBond	C390	C407
SSBond	C392	C409

We had expected these strong constraints to make folding easy, but we
were unable to get a really compact structure.  One problem is that
undertaker has few operators for rearranging a chain once the
disulfides have formed and gaps have closed.

To speed up the searching, we broke the target into 3 domains: 1-219,
215-339, 334-445.

This is try11-opt2, the best-scoring with an undertaker cost function
that has no constraints except the ssbonds.