curious to know whether only 3-helix binder protein could be design if using scaffold-based binder design.

[Huilin-Li]

Firstly I used de novo binder design --> only 1-helix binder protein will be designed. Then, I am using scaffold-based binder design by using the given 1000 scaffolds --> only 3-helix binder protein are designed. (inference.num_designs=10000).

Is it normal?

[roccomoretti]

Helical and helical bundle proteins are indeed favored by RFdiffusion, as it's slightly easier to diffuse into helices than beta sheets. However, RFdiffusion should be able to generate a large number of different topologies, so it's a bit surprising you don't see any variation over 10,000 designs.

My only thought is that if you have a particularly short protein, it may be hard for RFdiffusion to create a well-defined protein core with anything besides a short alpha helix or small helical bundle. You may want to try increasing your protein length (80-100 aa is often what's used for de novo design, but longer gives RFdiffusion more room to play around) to see if that increases the topological variation.