RFdiffusion

RFdiffusion copied to clipboard

I thought that if I add /0 in contig, that segment will become a separated chain in the result protein. for example ,
'contigmap.contigs=[A229-445/0 B230-444/0 20-20/C70-328/50-50]' , RFDiffusion will generate 3 chains. However I only get two chains in the result, Chain A and Chain B are merged.

I tried to separate Chain A and Chain B by a new contig(chain order is A C B): [A229-445/0 20-20/C70-328/50-50/0 B230-444/0 ], I still got the same result, i.e. target chains were merged to one chain.

I also tried a RFDiffusion wrapper from Sergey's colabdesign, I found he removed all the chainbreaks(/0), RFDiffusion only generates a single chain for both target and binder, then he uses 'run_mpnn_only.py' to separate this single chain to 3 chains.

I'd like to know if it is possible that RFDiffusion generates 3 chains based on provided chain break in contig.

Thanks Bin

I met this problem too, and solve it by modifying source code in model_runners.py and contigs.py

Hi, @GCS-ZHN. Would you mind to share how you modify the source code in model_runners.py and contigs.py in details? Since I meet the same problem too. Thank you :D.

Hi, @GCS-ZHN. Would you mind to share how you modify the source code in model_runners.py and contigs.py in details? Since I meet the same problem too. Thank you :D.

Modification is my forked repo， it will split inpaint chains as contig defined.

So I also ran into this problem. Also it seems like the downstream processing using this pipeline https://github.com/nrbennet/dl_binder_design is designed to only work with target and binder? Is there a workaround there? I am afraid that having all my chains fused into one target will interfere with alpha fold.

I've found this problem too. As a probably related issue, it would be good if RFDiffusion generated the chains in the same order that they are in the contig