At first we thought this might be a fold-recognition target, but
neither of our top two superfamily hits turned out to give us decent
alignments, and the lower hits were even weaker.

We tried to build the beta structure manually, but there were too many
strands and we got buried in the combinatorial complexity.
We did get a few bits and pieces of super-secondary structure, but
nothing that really looked like a full fold.

We have provided a scattered selection from the models we generated.

Model 1 is try17-opt2, best scoring with a cost function that has
constraints for predicted secondary structure.