In the process of trying to learn protein design for the last year and a half I have been focusing on using the newest and best tools for designing sequences. And the tools for designing a sequence to match a conformation seem to be reaching a level, such as with the protein MPNN release, that they're almost plug and play.
But I've come to realize that just because a backbone looks realistic doesn't mean it is. As a proxy to check this, when designing a sequence with either trRosetta or MPNN and then running it through a program like AlphaFold or RoseTTAfold I can see pretty quickly if the structure/sequence are in the ballpark or are totally unfoldable.
My question is this: how does one avoid the process or blind trial and error when trying to get a backbone that fits the needs of a project but is also in realistic conformation space? Is it truly a numbers game of eyeballing realistic structures and repeating? Or are there tools and metrics to do the grunt work prior to sequence design?
any answers would be greatly appreciated!