RosettaCommons
No cyclic backbones are generated with RFpeptide!
I am now trying to run RFpeptide on the Colab. I executed the command below, and the structure shown in the following photo was generated. I don't think that this command correctly generated a macrocyclic binder. Is there something wrong with this code, or doesn't this code always generate a ring-shaped backbone?
./RFdiffusion/run_inference.py inference.output_prefix=outputs/Sample1 inference.num_designs=1 inference.input_pdb=outputs/Sample1/input.pdb diffuser.partial_T=20 +inference.cyclic=True +inference.cyc_chains='A' ppi.hotspot_res=[B46,B48,B49,B50,B60,A63] 'contigmap.contigs=[15-15/A16-16/99-99 11-11]' inference.dump_pdb=True inference.dump_pdb_path='/dev/shm'
Out of curiosity, which CoLab notebook are you using? Can you respond with the link?
There are some known issues with the code not always producing cyclic peptides, even when running the provided examples. Issue 359 I think is the best to look at about this. If you're not using Stephen Rettie's notebook, you will likely encounter issues.
Thank you for your response. I use the Colab notebook provided by Sergey Ovchinnikov, and I added two codes on it:" opts.append("+inference.cyclic=True") opts.append("+inference.cyc_chains='B'")". The reason why I set cyc_chains to "B" instead of "A" is that the code always sets chains' name as chain "B" unintentionally. Actually, this is another problem I am facing now. Even if the target protein has chain "B" from the beginning, my code delete it and set the generated chain as "B"! But, I'll put it off for now...
I searched the notebook by Stephen Rettie but couldn't find. I would like you to post the link if you can.
I unfortunately cannot find a link to Stephen Rittie's CoLab notebook.
Have you tried running RFdiffusion outside of a notebook? If you are not running on ARM-based systems the Docker image allows you to bypass any dependency issues.