Open gatctg opened 8 months ago
I had exact the same issue KeyError: 'masked_list' , Hopefully get resolved soon.
Hi @gatctg @QUEST2179!
My apologies for late response due to being super busy these days. I have updated code for a unified and automatic data processing for both single and multi-chain structures. Please give it a try! You can also take a look into examples/inspect_data_and_model.ipynb
.
Hopefully this could be helpful! Thank you for you interest in our work again.
Best, Zaixiang
Hi Zaixiang,
Apologies for my delayed response due to end of year business as well. I am able to run both the single and multi chain structures, in examples/inspect_data_and_model.ipynb
, and they work beautifully, but am struggling with interpretation of the multichain output, especially when setting masked_chain_list
. My question is threefold:
masked_chain_list
, what is the interpretation of "remaining chains serving as conditioning?" I see that, by changing the chains specified in masked_chain_list
, I get back different predictions. However, none of the predictions contain a subset of the original sequence. Is it correct to ignore predictions of chains not included within masked_chain_list
, when specified?Thanks again for your previous help and eagerly await your reply.
Hello,
I have been able to successfully generate protein sequences for a custom backbone using
lm_design_esm2_650m
pretrained on thecath_4.2
dataset. However, when I try the same model pretrained on themultichain
dataset, I get the following traceback:when running the following:
On a similar note, is there a way to specify the chains you want to design in
multichain
? Sorry if I am missing something and thanks in advance for troubleshooting help.