Scaffold-based binder design performs much better than de novo binder design, and the scaffold has great influence in binder design

[Huilin-Li]

I am using RFdiffusion system to design binders for a while. I just find how important the scaffold is in binder design.

At the very beginning, I only use like design_ppi.sh to achieve de novo binder design. The process is pretty long and results are not good. Only one-helix binders are designed and hotspots does not work well.

Then, I tried like design_ppi_scaffolded.sh to design binders based on given scaffolds in ppi_scaffolds_subset.tar.gz. Designed binders become diverse, but the diversity of binders is still pretty limited (mainly 3-helix). At the same time, I also have to say, I still do not think the hotspots do work. Designed binders are still pretty random placed around the target. Maybe, there is a set of weights in front of these hotspots and we could increase these weights and strongly make RFdiffusion to design binders around the hotspots.

Currently, I get a protein from my colleague, and this protein is the binder they expect. I use this protein as scaffold to deign binders. Most of things become better and reasonable.

Any other suggestions on binder design with much better pae_interaction score? What other else I need to check out? It is pretty interesting to figure out one question step by step 😆 Thanks in advance for any suggestions and recommendation!