Free Energy Methods (FEP)

Background

Achieving highly potent binding, while maintaining a host of other ligand properties required for safety and biological efficacy, is a primary objective of small molecule drug discovery. During the past decades, free energy calculations have benefited from improved force fields and sampling algorithms, as well as the advent of low cost parallel computing. Such developments have enabled free energy calculations to yield useful comparisons with experiment, and play a role in academic drug discovery efforts. However, it has been challenging to develop free energy calculation protocols that achieve the accuracy, reliability, ease of use, and throughput that would be needed for such calculations to impact lead optimization in an industrial setting.

 

Seeing this clear unmet need, we embarked on a multiyear research project, in tight collaboration with our Scientific Advisory Board, to develop a new free energy calculation technology (FEP+). Our objective was to provide a rigorous approach for computing binding free energies that provides significant value to industrial drug discovery efforts.  We are pleased to report that, after utilizing the FEP+ technology on 7 different active drug discovery collaborations over the past year, we now have firm evidence that the free energy approach developed in FEP+ can facilitate better synthesis decisions during lead optimization.1

In Figure A, we report the results of all 138 experimentally verified prospective predictions across all 7 active drug discovery collaborations available at the time of this writing. In 63% of cases, the FEP+ affinity predictions are within 1 log unit of the experimental value, and in only 6% cases are large prediction errors (> 2 log units) observed; see Figure B.

Figure A - The results of 138 FEP+ prospective predictions, spanning 7 active drug discovery collaborations with industrial partners, are plotted against experimental pIC50; the dark gray band represents 0.8 log unit of error in the predictions and the light gray band represents 1.6 log units of error in the predictions.

Figure B - A histogram illustrates the percentage of predictions that fall within each mean unsigned error (MUE) bin.

Figure C - A truth table summarizes the results of the FEP+ classification of a molecule being a tight binder (< 100nM).

Conclusion

This enrichment of synthesized tight binders compares quite favorably with industry averages, where we estimate only 10-40% of molecules synthesized during lead optimization will typically have sufficient potency to plausibly become development candidates.2 Furthermore, we estimate at the time of this writing, approximately 1,500 idea molecules across these 7 projects were deprioritized for synthesis on the basis of FEP+ calculations. Those deprioritizations likely both avoided the potentially costly synthesis of many weak binders, and perhaps more importantly, accelerated the rate at which more promising tighter binding molecules were discovered over the course of these active projects.

 

The achievement of these results is a consequence of multiple technology advances including:

  1. Completion of the OPLS3 force field3, which incorporates refined van der Waals parameters and partial charges, as well as >10,000 highly optimized torsional parameters for drug like molecules
  2. Development of the FFBuilder, which can­­—in a completely automated fashion—fit any unparametrized torsions that may exist for the idea molecules of interest
  3. Incorporation of the REST enhanced sampling methodology, which enables detail-balance-preserving simulation of a selected subsystem at a higher effective temperature, thereby focusing sampling efforts where needed to most efficiently traverse the relevant phase space4-7
  4. Optimization of the FEP+ algorithm to make use of the efficient Desmond GPU dynamics engine8, which enables 4 FEP calculations to be completed per day on a relatively inexpensive8 GPU server
  5. Development of the FEP Mapper user interface, which enables calculations to be set up and run with only minutes of user intervention, and further facilitates a deep probing of the consistency and convergence of the simulation results.7,9

 

As the FEP+ technology continues to mature, we see great opportunities to broadly deploy the technology in a variety of settings to aid the more rapid discovery of promising chemical matter. Among the greatest challenges in lead optimization is the relatively slow rate at which new molecules providing critical insights become available to the project team. During lead optimization, often only 20 or fewer molecules are synthesized per week, yielding around 1000 data points over the course of a year. This can make the synthesis queue the single greatest bottleneck in the acceleration of the discovery project. FEP+ scoring can thus help to maximize the value of each round of synthesis by helping to ensure a greater fraction of compounds will have the necessary potency to meet project goals. We further expect such potency-guided discovery to facilitate the more aggressive optimization of many other ADMET properties essential to project goals with greater confidence that compound potency will not be compromised, and embolden project teams to pursue new synthetic directions that would have been considered too risky without the increased confidence provided by FEP+. Thus, we firmly believe the FEP+ technology is well positioned to enable project teams to discover development candidate molecules more efficiently and successfully than would otherwise be possible.

 

References

  1. Wang, L. et al., Accurate and Reliable Prediction of Relative Ligand Binding Potency in Prospective Drug Discovery by Way of a Modern Free-Energy Calculation Protocol and Force Field. J. Am. Chem. Soc., 2015, 137(7), 2695–2703.
     
  2. Hann, M., Molecular obesity, potency and other addictions in drug discovery. Med. Chem. Commun., 2011, 2, 349-355.
     
  3. Harder, E. et. al., OPLS3: A Force Field Providing Broad Coverage of Drug-like Small Molecules and Proteins. J. Chem. Theory Comput., 2016, 12(1), 281-296.
     
  4. Wang, L.; Berne, B.J.; Friesner, R.A., Ligand binding to protein-binding pockets with wet and dry regions. Proc. Natl. Acad. Sci. U.S.A., 2011, 108(4), 1326-1330.
     
  5. Liu, P.; Kim, B.; Friesner, R.A.; Berne, B.J., Replica exchange with solute tempering: a method for sampling biological systems in explicit water. Proc. Natl. Acad. Sci. U.S.A., 2005, 102(39), 13749-54.
     
  6. Wang, L.; Berne, B.J.; Friesner, R.A., On achieving high accuracy and reliability in the calculation of relative protein–ligand binding affinities. Proc. Natl. Acad. Sci. U.S.A., 2012, 109(6), 1937-1942.
     
  7. Wang, L. et al., Modeling Local Structural Rearrangements Using FEP/REST: Application to Relative Binding Affinity Predictions of CDK2 Inhibitors. J. Chem. Theory Comput., 2013, 9(2), 1282–1293.
     
  8. Bergdorf, M.; Kim, E.T.;, Rendleman, C.A.; Shaw, D.E., Desmond/GPU Performance as of November 2014. DESRES/TR--2014-01.
     
  9. Liu, L.; Wang, L.;, Mobley, D.E., Is Ring Breaking Feasible in Relative Binding Free Energy Calculations? J. Chem. Inf. Model., 2015, 55(4), 727–735.

Read more science articles

Back To Top