This is the best we could in the limited time.  

We do not like the two antiparallel strands at Q249-A266---these were
picked up from one of the templates, but we believe that this region
should be a helix.  Some of the other secondary structure predictions
also don't match--probably indicating other alignment errors.

This model was created by taking the fold-recognition results, adding
strong constraints for the secondary structure prediction, optimizing with
undertaker, then using VAST to get a new alignment to the penteins
from the predicted structure.  The process was iterated, but it is
clear that more knowledge (or lucky guesses) of the sheet structure is
needed to get a good alignment---some parts of the structure need to
be modeled ab initio.

We also included some constraints to try to keep the conserved polar
and cysteine residues together, believing these to be part of the
active site.  It isn't clear whether these constraints were needed.

The final model had the sidechains repacked by Rosetta---for once the
repacked version scored better with the undertaker cost function than
the undertaker-only model.  This may have been due to the very short
optimization runs we did with undertaker, due to tight deadline pressure.