search

veghen / ProRefiner

Code for ProRefiner: An Entropy-based Refining Strategy for Inverse Protein Folding with Global Graph Attention
4 stars 1 forks source link

ProRefiner change more than desired residues while doing partial design. #3

Closed DaHenz closed 9 months ago

DaHenz commented 9 months ago

Hi, I tried to test a partial design with the ProRefiner, whereas I want to design one AA in one chain from a pdb. Interestingly, in the output from MPNN, the desired residue was changed, But in ProRefiner two residues were changed.

python run.py input/test8.pdb B 7

Recovery and nssr are computed based on designable residues with coordinates only.

Design 1 residues from input/test8.pdb chain B with length 15

native sequence: GSARQENRKEARQEG

sequence by ProteinMPNN: (recovery: 0.000 nssr: 0.000) GSARQETRKEARQEG

sequence by ProRefiner: (recovery: 0.000 nssr: 100.000) GSARQESRLEARQEG

veghen commented 9 months ago

Sorry for the mistake. The model performance is evaluated only on designable residues, so the non-designable part was initially overlooked in ProRefiner's output and it doesn't influence the evaluation results. I have fixed this by replacing the non-designable part with ground truth sequence in the output. Please check the latest commit. Sorry again and thank you very much for pointing this out.