Thanks for the response, huxz.
I looked through some of the scoring function code yesterday, and I think I might maybe kinda sorta possibly understand a little better how to do this now… It looks like when designing proteins de novo you want to only go through the first few stages of scoring (bump check first, then ss & light strand pairing, etc.), but not the later stages where centroids are added in for the side chains and the chain is collapsed? That way you get backbone models “generated without regard to side-chain packing” (Kuhlman et al., describing Top7), right? I was confused before because I was worried that if I put in poly-Ala or something as my sequence that the starting models would be packed too tightly to allow mutation to larger side chains later. But if you cut out the later scoring stages that take side chains into account (the sequence-specific part of the scoring function), then this isn’t a problem.
Is that correct?