baker-laboratory / rf_diffusion_all_atom

Public RFDiffusionAA repo
Other
288 stars 44 forks source link

Question About Sequence Design Evaluation #8

Closed warriors44 closed 6 months ago

warriors44 commented 6 months ago

Hello, and thank you for developing such useful software.

I'm not sure if this is the appropriate place to ask, but I have a simple question regarding the design method described in your paper, specifically about the use of RFdiffusion-AA. After designing sequences with Ligand/ProteinMPNN, I noticed that RoseTTAfold-AA was not utilized to evaluate the validity of these sequences. Is this because you believe that the Rosetta relax and the AF2 method provide a more reliable evaluation?

Thank you.

ikalvet commented 6 months ago

We used AlphaFold2 to have an independent evaluation of the sequences, so that the selection process wouldn't be biased by the method that produced the backbones.

warriors44 commented 6 months ago

I see. That makes sense. Thank you.