issues about cyclic peptide generation

[hahadg]

Hi developers,

I have successfully installed CycleDesigner on my system, but I have some questions regarding cyclic peptide generation.

First, in the rfd_zch_inference.sh script:

INFERENCE_SCRIPT=./scripts/run_inference.py  
INFERENCE_MODEL_DIR=/home/yons/models

Do these refer to the run_inference.py script and model from RFDiffusion? I set INFERENCE_SCRIPT and INFERENCE_MODEL_DIR to the paths of the corresponding script and model from RFDiffusion. However, when I tested it using the example PDB 6jwm from the paper, the generated peptide was linear rather than cyclic. This seems no different from using RFDiffusion directly.

Also, the generated sequence consists only of glycine residues. Does this mean I need to use ProteinMPNN to design the actual sequence, and then use HighFold to predict the structure?

Looking forward to your response!

[min0210]

I'm having the same problem.
My result is linear peptide.
Shouldn't we use cyclic=True option in inference?
@zzzcchhw @hongliangduan