T0199 was split into three regions: domain 1 (1-95), domain 2
(131-end), and a linker (81-136).  The two domains were separately
modeled with our usual methods, then we tried pasting them back together.
We had a lot of difficulty with this, as the simple superposition on
our best alignment model made the domains overlap badly.  Eventually
we added constraints to the linker region until it was reasonable
(though the domains got destroyed in the process), then superimposed
the domains on the linker.  

Model 2 is the result of doing further optimization after the
superposition.  The linker straightens into a single helix, but
collapses down against domain 2, so that the domains end up closer
together.  We don't like this solution as much as the simple
superposition, but some of the details in the domains are better.