How to use the trb file in combination with ProteinMPNN

[wangdoc2002]

Hi, RF diffusion team, I am inpainting a protein with varied AA length. For a fixed length of design, I can use --fixed_positions_jsonl function in ProteinMPNN.

For a varied length of designs, the .trb file outputs which residues were fixed during RFdiffusion nicely. Is there a way I can fed this trb information to ProteinMPNN?

[joewatchwell]

Yes! So the trb file contains the mappings of where residues were in the input, and where they ended up in the output. These are "con_ref_idx0" and "con_hal_idx0" (named such for historical reasons). Everything in "con_ref_idx0" are residues that were fixed in the input contig. "con_hal_idx0" contains the corresponding 0-indexed position of those residues in the output. This can then quite easily be used to then make the fixed_position dictionary inputs to ProteinMPNN. We have scripts for doing this - if you have any difficulties we can probably share these

[noah-c-noah-do]

@joewatchwell that script would be very helpful! Could you share it? Many thanks!

[BaylenRavenscraft]

If that script is available, can you please share a link? Thanks either way!