For T0198, we had fold-recognition alignments to various all-alpha
proteins, but none of them were a convincing model for the whole
protein, so we ended up folding this one pretty much from scratch.

We ended up with two main hypotheses:
	an up-down bundle
	an alpha sandwich

Our best-scoring model with cost function that includes the predicted
helices is try20-opt2, one of the bundle models.  
It also breaks predicted helix P76-M108, but we are submitting it as
model 2.