The basic fold for T0211 was solidly established by homology, but
there was some question about residues E86-H95, which our neural nets
predicted to be a helix, but which aligned to a strand in several of
the templates. Although we got a helix fairly easily, getting it
oriented correctly took some work with constraints to force it out of
the position it initially adopted.

We also had problems with the buried charges E14 and K76, finally
deciding to make a salt bridge joining them.

This model (try17-opt1) scores best of the models we generated with an
undertaker cost function that had just the salt bridge as a constraint.