Unbinding Kinetics Best Practices

This document describes best practices derived from the deployment of Unbinding Kinetics workflow in drug discovery projects conducted by Schrödinger Inc., and how issues with the workflow are identified and addressed.

Before You Begin

A comprehensive target analysis is undertaken to evaluate the appropriateness of a target for structure-based drug discovery methodologies (e.g., Glide, WScore, WaterMap, FEP+) before proceeding to unbinding kinetics calculations. To perform this target analysis:

1. Identify all available crystal structures of the target, and analyze crystallographic data to gauge the binding site's propensity to reorganize upon ligand binding.

2. Refine ligand-protein co-crystal structures using PrimeX, consult with in-house crystallographers, and ensure the structures are complete (no missing residues or atoms) and consistent with the electron density. Model in any missing atoms or residues. All crystallographic water molecules should be retained and included in the X-ray refinement, and crystal waters not overlap with the ligand(s).

3. Identify several compounds with known affinity and kinetics data; ensure there is a quality co-crystal structure with one of these ligands or a similar ligand.

4. Confirm that the experimental methods used to measure the affinity data can be unambiguously related to the binding free energy and that the crystal construct has a one-to-one agreement with the protein construct in the assay. It is critical to thoroughly understand the details of the assay, e.g. if the protein is a monomer or dimer in the assay, etc. It is also critical to understand the pH, ionic strength, denaturing/stabilizing agents, added ions, temperature, and solvent conditions of the assay. If these deviate significantly from the conditions of the simulations or the crystallography, systematic errors may emerge.

5. Determine the best possible binding modes for the ligand(s) with known binding affinity using all available information, including all relevant protomers and tautomers, possible flipped states of asymmetric rings, etc.

6. Run the Force Field Builder to parametrize any missing torsions for the ligands. See the Force Field Builder - OPLS4/OPLS5 Panel for more information.

7. Run FEP+ on the ligand(s) and compare the predicted ∆∆G values with experimental ∆∆G values (for relative binding FEP+), or compare the predicted ∆G values with experimental ∆G values (for absolute binding FEP+).

   1. The Bennett errors should be less than 0.3 kcal/mol

   2. The RMSE of FEP jobs should be less than 1.3 kcal/mol

   3. There’s no evidence of convergence problems requiring further investigation

   4. Proceed to the next step

8. Run the Unbinding Kinetics workflow on the ligand(s) with known residence times, and compare the predicted with the experimental values. Ensure the criteria below are met before proceeding to the prospective studies (jobname-multisim.log will have all the information).

   1. Predictions should be within 1-2 orders of magnitude (OoM).
   2. Most of the RAMD jobs should be completed within 1-2ns

Sources of Error

The following are potential sources of error for the Unbinding Kinetics workflow:

1. Inadequate sampling or high ligand flexibility

   • Ligands that are too flexible within the binding pocket
   • Ligands with many degrees of freedom (see Out of Domain of Applicability)
   • Flexible proteins are known to undergo large conformational changes during unbinding events (see Out of Domain of Applicability)

2. Too fast egress during the RAMD stage when using the default force (K) of 15 kcal/mol/Ų

   • RAMD simulation time < 100 ps
   • Experimental residence time < 10 s
   • Likely to occur when simulating fragments

3. High standard errors in the RAMD stage

   • Standard error >= 50% of the mean RAMD time

The above RAMD thresholds are from Koch et al. Schrödinger 's Unbinding Kinetics worflow is more sensitive so the tuning of parameters may happen earlier.

Troubleshooting

The following steps could be taken to address the sources of error mentioned above:

1. Inadequate sampling or high ligand flexibility: When dealing with highly flexible systems or when the ligand is not stable within the binding pocket, consider increasing the equilibration time (-stability-sim-time)

2. Too fast egress during the RAMD stage when using the default force (K) of 15 kcal/mol/Ų: Typically, μM to nM ligands should run between 100 ps and ns and cause no problem. If this is not the case, or more likely if dealing with fragments, consider reducing the RAMD force (-ramd-spring-fk) to K= 10 or 5 kcal/mol/Ų

3. If the standard error is >= 50% of the mean RAMD time, consider increasing the number of RAMD replicas (-total-ramd-replicas)

Exit Statuses

Each metadynamics replica reports a status in the job log (jobname_7-out/job.log). The possible statuses are:

• OK—The ligand successfully unbound from the receptor.

• Simulation Timeout—The maximum simulation time (default 200 ns) was reached before an unbinding event occurred. This does not necessarily indicate a problem with the collective variable. You can recover the correct statistics by increasing the simulation time cutoff (-time).

• Residence Timeout—The residence time cutoff (default 10¹² s) was reached. This indicates that the system may require further investigation.

Out of Domain of Applicability

The following are cases out or on the edge of the domain of applicability:

1. Floppy and highly flexible ligands (e.g., long peptides)

2. Unbinding events that require receptor reorganization

Support

For questions and support, please contact the following team leads, or an Application Scientist:

• Schrödinger Unbinding Kinetics Development Team Lead: Dmitry.Lupyan@schrodinger.com

• Schrödinger Force Field Development Group Lead: Ed.Harder@schrodinger.com