Learn/Redesign a sequence
Score every residue
Ask the model about the sequence that is already there. For each residue ProteinMPNN gives a probability to every amino acid, given the backbone and all the other residues. The probability it gives the one present is the residue's native fit, written as a log, so 0 is a sure fit and the more negative the less the model expected it; how peaked the distribution is, its certainty. The mean fit over the complex is the sequence's score: -1.37 here.
Chain B by native fit
Loading structure
- structure
No sequence loaded.
The ink marks what to look at: a residue the model would not have chosen there. The Certainty tab shows where it had no clear preference at all.
Read more: Phi API: per-residue probabilities · Notin et al. 2023, ProteinGym