Docking + Active Site Design

I just responded to your other post – https://www.rosettacommons.org/comment/8166#comment-8166

Regarding enzyme design, there really isn’t anything special about a “transition state” in enzyme design, as far as Rosetta is concerned. It’s just another ligand. It may be chemically unrealistic, but with the correct input settings that’s not much of a concern to Rosetta. At that stage in the process, Rosetta is functioning almost solely on the transition state stabilization theory for enzyme activity, so it’s not a big leap to change the protocol slightly in order to design for binding of a chemically realizable compound instead of a transition state. In fact, it’s a little bit simpler as you don’t have to worry about the active site catalytic constraints. (Although you can certainly put them in for binding interactions instead, if you want.)

See Tinberg et al. Nature. 2013 Sep 12;501(7466):212-6. for an example of using the enzyme design machinery to redesign for small molecule binding.

Generally for ligand docking and enzyme design the binding energy of the ligand to the protein (roughly interface_delta_X) is the primary energy for optimization. The reason for this is that there can be a lot of noise in the total energy due to movement of protein sidechains which wouldn’t affect the binding energy at all.

The caveat to this is if you go just on the binding energy, you can theoretically cause the protein to unfold or make clashes with itself to get a better interaction with the ligand. You want to avoid this, so often there’s a multi-stage analysis involved, where you first do a filter on total score to throw out all the bad-scoring outliers (which may have clashes or other bad geometry.) From these reasonable structures you then rank by interaction energy and pick the structures which best optimize it.

The “select_best_unique_ligand_poses” application (https://www.rosettacommons.org/docs/latest/ligand-dock.html#Clustering-ligand-poses) is probably my recommendation.

This doesn’t do full-on clustering in the K-means naive sense, but instead takes an energy-based approach. It takes the lowest energy ligand, and labels that as a cluster center. It then runs up the list of structure sorted by energy until it finds a structure which has a ligand rmsd outside the provided cutoff radius, and then labels that as a new cluster center. It continues this until it’s found a given number of cluster centers, or it exhausts the number of structures available.

If you use the version in a weekly release, it should take the standard “jd2” option for input.

One drawback is that it’s set up just to give you cluster centers, not to say which cluster non-centers are part of. It should be easy enough to alter the program to spit that information out, though.