Aleksey Murzin suggested on FORCASP that the main difficulty with this
target was choosing between a monomeric and dimeric model:

  "It is predicted that an "un-swapped" monomer may adopt a similar
   fold to the distantly related monomeric TolA domain (1TOL,
   1LR0). It probably retains the conserved TolA/TonB structural core
   but the rest of its chain may fold differently. With two molecules
   in the asymmetric unit, is the target structure the dimer or a
   monomer? The choice is yours."

We built both monomeric and dimeric models.  The dimers were built
starting from the 1ihr dimer.  Since undertaker does not currently
support optimization of dimers, we made a linked dimer by copying the
target chain twice and optimized that.  The resulting double-length
monomer was badly distorted in the region of the linker, so we
superimposed the first and second halves and made a chimeric monomer
from the N-terminal region of the first half and the C-terminal
region of the second half.  This monomer was dimerized by
superposition on 1ihr, and another double-length optimization done.
Another crossover was done, and optimization was done with a forced
break.  The final dimeric model had both monomers almost identical.

The monomeric models were built by taking one of the dimer models (not
the final one, but one iteration earlier) and cutting and pasting from
the two chains to make a compact monomer with a few bad breaks.  The
residues N51-K82 were taken from the "other" chain.  This "folded
monomer" was then optimized to close gaps and improve packing.  The
resulting monomer was not quite as densely packed as we would like, as
improving the packing would require bending the sheet more, which
requires simultaneous changes in three strands---a conformation-change
that undertaker is not currently capable of.

Because of the need to curl the sheet more in the monomer, and because
turn at K82,I83,N84 looks more like the residues one would expect 
mid-strand, we favor the dimeric model over the monomeric one.

This is try17-opt2, our best monomeric model.
It still has some exposed residues that should be buried.