I am interested in using AlphaFold2 to identify residues within a homodimer that are important for the protein-protein interaction. I have been able to use ColabFold to produce a predicted dimer structure - however, as far as I can tell AlphaFold does not actually identify the residues or regions that are actually important for binding. Is there a way to manipulate the software to do this? Thanks!
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
I don't think that the full workflow lies inside alphafold. However, you can try a workflow like this:
- identify interface residues (e.g. in pymol)
- mutate each interface residue in the dimer structure sequentially. There are various utilities that can do this, I believe that pdbfixer can do it.
- minimize each resulting mutated dimer (use openmm or fastrelax)
- estimate stability of mutants using AF2 functionality
If you compare the stability of each mutant to the wild type dimer structure, you will get some sense of how important each interface residue is to the AF2-estimated dimer. This will not be perfect but you can likely roughly gauge residue importance in this fashion.
-
1$\begingroup$ All good points. I would add if you are going for a more high-throughput method you can also use metrics like pDockQ, and the self-reported ipTM values to estimate how correct the model has predicted the interaction. From my experience, and especially with smaller molecules, I notice that the orientation of the complex can change quite easily from model to model. My point is that is that if the model is poorly constructed, you might need another strategy. $\endgroup$– dthorburCommented Apr 4 at 14:28
-
$\begingroup$ @dthorbur that's a great point. my suggested strategy depends on the accuracy of the pose in the first place, so ideally you'd have several starting points. $\endgroup$ Commented Apr 4 at 16:27