How to generation many seq in one backbone

[didiwei1987]

hello, nice work. when i use it to design protein binder, I find it had generate sequence not GGGG...... Is it necessary to continue executing ProteinMPNN for sequence generation as well as structure prediction using RF3?

[Ubiquinone-dot]

Hello,

Do you mean that the designed binder comes out from RFD3 with non-canonical amino acid assignments? Could you send me your arguments to rfd3 and I would be happy to help debug!

[rclune]

Were you expecting an output like RFdiffusion (the original) where all of the residues in the designed portions were glycines? RFD3 operates on the atom level instead of the residue level, so the outputs now have designed sequences as well. You should still be able to feed these outputs into sequence design tools like LigandMPNN/ProteinMPNN.

[JA-Hull]

I had the same thoughts as didi, I don't see any metrics like iPAE etc so I assume going to RF3 is required at minimum to get that. But I can't help but wonder if other auxiliary potentials like physics based scoring might make RF3 redundant.