search

dauparas / ProteinMPNN

Code for the ProteinMPNN paper
MIT License
910 stars 278 forks source link

Clarify 'design chains' #78

Open jadolfbr opened 8 months ago

jadolfbr commented 8 months ago

In the examples, one gives 'design chains' to ProteinMPNN. However, when you give specific positions to design or not design, this gets odd.

For example, if I have a complex of A B C, and I'm designing chain B. I give chain B and a set of non-fixed residues. However, the design of chain B does not take into account A and C. IE design chains is something more to do with masking. If I give A B C, then MPNN takes A and C into account when designing chain B.

I think this is a bit counter intuitive - perhaps some updates in the docs/examples would be welcome here.

proteins247 commented 6 months ago

I had a complex of A B C D. I tried designing specific residues in chain D (like submit_example_4_non_fixed.sh). Given your post, I tried out the following

I found that the residues of the two experiments were very similar, suggesting that the program is accounting for the other chains.