Usage question

[ahof1704]

Hi,

I have a basic question about the usage of PepMLM. I would like to design a binder for a target protein sequence that is ~20aa. I would like the designed protein to be around ~80aa long.

I have tried to use PepMLM to design the full 80aa-long protein, but it was mostly linear and it doesn't seem to fold well.

Would you recommend a different setup, such as designing a much shorter binding peptide to my target sequence followed by some scaffolding using a different method (eg, RFdiffusion)?

Thanks in advance

[ahof1704]

Could anyone help with the question above? Thanks

[programmablebio]

Hi! Yes, I would suggest not making a binder directly of that length. If you take a look at the paper and the datasets used to train, you will see that our training set did not have binders that long. Plus, it's a peptide generation model, not a miniprotein model. I would use PepMLM to design a ~20AA peptide and then scaffold it with RFDiffusion, as you mentioned.