Preparing Protein and Ligand Structures for FEP+

Protein Structure Preparation

Protein model completeness: Protein preparation should include fixing any chain breaks, modeling in any loop conformations and adding any missing side chains. Chain breaks near the active site will likely lead to poor results. Disulfide bridges should be created and termini residues capped where applicable.

Crystallographic waters: Include all crystallographic waters that do not have severe clashes with ligands in the protein entry. For example, there is a water-mediated interaction between the protein and the co-crystallized ligand in the BACE1 Inhibitor Design Using Free Energy Perturbation tutorial (4DJW_4j_prepared) and so it should be included. Sometimes WaterMap hydration sites can be added to fix waters missing from crystal structure.

Orientation: The protein can rotate during the simulation, which has the potential to cause undesirably large self-interactions or even clashes. This is especially the case if the protein is elongated. To reduce the likelihood of these occurrences, it is recommended that the protein is aligned so that the vector between the two atoms that are separated by the largest distance is aligned along one of the coordinate axes. You can choose View → Align View to perform the alignment (change the view, see Aligning the View to the Structure); click Update Coordinates in the panel after alignment to change the coordinates. (It does not matter if the two atoms you pick are not exactly the furthest apart, as long as they are close. If you want to check, you can measure the distances.)

Ligand Preparation

LigPrep: Careful preparation of the ligands is critical to a successful FEP+ prediction. Best practices include running LigPrep on all the compounds to exhaustively enumerate all the stereoisomers and likely protonation states of the ligands. Note that triply-substituted ammonium cannot invert stereochemistry during the simulation, making it important to model both pseudo-stereoisomers.

Jaguar pKa: It is especially important to model the correct protonation state of the ligand. Epik state penalties will be returned from a default LigPrep run and these should be followed up by Jaguar pKa calculations for any site whose protonation state is ambiguous. See Jaguar - pKa Panel for more information.

Force field coverage: Force field coverage of torsional angle space should be assessed before running FEP+. This can be evaluated directly from Maestro using the Force Field Builder - OPLS4/OPLS5 Panel. Please see Force field coverage for more details.

Initial conformation: Given infinite sampling, the results of the FEP+ job would not be sensitive to the initial conformation of the ligands. However, to achieve convergence and reliable results during the default 5 ns simulation, two things are crucial:

  • The initial conformation must be as close to the native binding mode of the ligand as possible.
  • The atoms that are common between two ligands being perturbed are essentially coincident in space.

For a single-atom perturbation, it may be easiest to duplicate the lead compound in the project table, and just edit the new ligand to make the change.

For small to medium perturbations, it may be sufficient to do an alignment in Maestro, such as in this tutorial. Visually inspect the poses to ensure there are no steric clashes with the protein. Docking ligands using Glide with core constraints can also be useful here.

For larger ring additions, MM-GBSA refinement of an alignment may give the best initial pose prediction. For example, initial ligand core alignment as above followed by an MM-GBSA sampling of the ligand R-groups usually provides a good initial pose for FEP+.

Asymmetrically substituted rings: FEP+ utilizes a molecular dynamics simulation to allow ligand movement in a binding pocket as the initial compound is perturbed to the target compound. However, larger groups - such as rings - will not likely have complete rotational freedom during the time scale of the simulation. As LigPrep evaluates ligands independently of the binding pocket, the orientation of asymmetrical rings that are generated in the LigPrep output may not be reflective of the best predicted binding pose and the ring may not be able to change its orientation during the FEP+ calculation. If the ring conformation is not known a priori, then it will be best to ensure that any FEP+ calculations will completely annihilate the ring as the compound is perturbed from one ligand to another. This can be achieved by adding in additional intermediates that do not contain the ring, or by using custom cores, or including both ligand conformations in the FEP map and using the pose with lower predicted free energy for the analysis.