For refining tr370, we tried both monomeric refinement and dimeric refinement. In the end, all the models that we are submitting have been refined as dimers. Our basic refinement protocol consisted either of repeatedly optimizing with undertaker, or of optimizing the model using undertaker, then reoptimizing with gromacs, and repacking the sidechains (except CYS and PRO) using rosetta. For each iteration of this protocol, we changed the undertaker cost function, to emphasize those features of the model that we felt needed the most work. This protocol alone did not seem adequate for getting the C-terminal sheet cleaned up, so we also tried making a chimera, using the refinement model for M1-A132, but taking W133-D154 from a model we had generated independently, which had better H-bonds for the final strands. Refinement model 1 is try18-opt2.unpack.gromacs0.repack-nonPC chain A, which had the lowest Rosetta energy of any of the dimer refinement models that Rosetta repacked sidechains for. It was optimized by undertaker from dimer/try17-opt2.unpack.gromacs0, then reoptimized with gromacs and with sidechains repacked by rosetta. dimer/try17-opt2 was optimized by undertaker from dimer-from-try73-opt2-1rfeA, which was a superposition of the monomer try73-opt2 on the 1rfe dimer. The monomer try73-opt2 was optimized by undertaker from try72-opt2 (no gromacs or rosetta), which in turn was optimized by undertaker from chimera64-46-no2, the chimera of try64-opt2 and try46-opt2. Try46-opt2 was a model we had generated ourselves (not derived for tr370). try64-opt2 < try63-opt2 < try62-opt2 < try55-opt2 < try53-opt2 < try52-opt2 < tr370.gromacs0.repack-nonPC. That is, the tr370 model was optimized with gromacs and had sidechains repacked by rosetta, then six runs of undertaker were done before the chimera was made. Refinment model 2 is chain A of dimer/try23-opt2.unpack.gromacs0.repack-nonPC, which has dense packing and a good rosetta energy (though not as good as model 1). This model was built to try to get better burial for V41, L43, W133, and L139, which seemed to exposed in try18. dimer/try23-opt2 < dimer/try22-opt2.unpack.gromacs0.repack_nonPC dimer/try22-opt2 < dimer/try21-opt2.unpack.gromacs0 dimer/try21-opt2 < dimer/try20-opt2.unpack.gromacs0 dimer/try20-opt2 < dimer/try17-opt2 Refinment model 3 is chain A of dimer/try12-opt2.unpack.gromacs0.repack-nonPC, which had the lowest Rosetta energy of any of the non-chimeric models optimized directly from tr370. dimer/try12-opt2 < dimer/try10-opt2 < dimer/try9-opt2 < dimer/try8-opt2 < dimer/try7-opt2 < dimer-from-try56-opt2-template-1rfeA The monomeric try56-opt2 was optimized by undertaker from try54-opt2 < try53-opt2 < try52-opt2 < tr370.gromacs0.repack-nonPC For dimer/try12-opt2.unpack.gromacs0.repack-nonPC, optimization by gromacs and repacking by rosetta ocurred only at the beginning (on tr370) and at the end (on dimer/try12-opt2). We probably could have gotten faster convergence had the cycling protocol (undertaker, gromacs, rosetta) been used, since gromacs is good at removing small clashes. Refinment model 4 is chain A of dimer/try17-opt2, which scored the best of the dimeric models with some of undertaker's cost functions, including the unconstrained costfcn and try23.costfcn. Its provenance was described above, as this model was on the path to both try18 and try 23. Refinement model 5 is chain A of dimer/try21-opt2, which scored the best with the try22 costfcn. It is on the path to try23-opt2.