Free Energy Perturbation (FEP) in Lead Optimization

What Is Free Energy Perturbation?

Free Energy Perturbation (FEP) is a rigorous computational method for predicting the relative binding free energy (ΔΔG) between two molecular analogs. Rooted in statistical mechanics, FEP calculates the free energy difference by alchemically transforming one molecule into another through a series of intermediate states, typically using molecular dynamics (MD) simulations.

In drug discovery, FEP is used during lead optimization to prioritize which chemical modifications will improve binding affinity, potentially replacing costly synthesis and testing of every candidate. The method gained prominence with the introduction of FEP+ by Schrödinger, which made FEP calculations practical for drug discovery through enhanced sampling and GPU acceleration.

Data: FEP Performance and Impact

A landmark study by Wang et al. (2015) demonstrated that FEP+ achieved an RMSE of 0.8 kcal/mol across 330 relative binding affinity predictions spanning 8 protein targets, establishing FEP as a reliable tool for lead optimization.

How: The FEP Calculation Pipeline

Step 1: System Preparation

1. Obtain protein-ligand complex structure (X-ray, cryo-EM, or AlphaFold 3)
2. Parametrize both ligands using OPLS4 or similar force fields
3. Define the alchemical transformation path (atom mapping between ligands)
4. Set up the thermodynamic cycle: ΔΔG_bind = ΔG_unbound - ΔG_bound

Step 2: Lambda Window Setup

1. Divide the transformation into 10-20 lambda (λ) windows
2. Each window represents an intermediate state: λ=0 (molecule A) to λ=1 (molecule B)
3. Use replica exchange with solute tempering (REST) for enhanced sampling
4. Position restraints on protein to maintain binding pose

Step 3: MD Simulation

1. Run MD simulations at each lambda window (typically 5-20 ns per window)
2. Use GPU-accelerated MD engines (DES-AMOEBA, GROMACS, AMBER)
3. Monitor convergence via overlap matrices and time-series analysis
4. Total simulation time: 100-500 ns per FEP calculation

Step 4: Free Energy Calculation

1. Apply the Multistate Bennett Acceptance Ratio (MBAR) or BAR estimator
2. Calculate ΔG for bound and unbound legs
3. Compute ΔΔG = ΔG_bound - ΔG_unbound
4. Estimate uncertainty from replica variance

Step 5: Decision Making

1. Compare predicted ΔΔG with synthesis cost and expected improvement
2. Prioritize compounds with predicted ΔΔG < -1.0 kcal/mol (significant improvement)
3. Consider SAR trends across multiple FEP calculations

Comparison: FEP vs. Other Binding Affinity Methods

Summary: Key Takeaways

1. FEP provides the most accurate computational binding affinity predictions, with RMSE approaching experimental uncertainty.
2. The method is best suited for lead optimization of closely related analogs, not initial screening.
3. FEP+ by Schrödinger has made FEP practical through GPU acceleration and enhanced sampling.
4. FEP-guided lead optimization can improve hit rates 2-5x compared to intuition-based design.
5. The accuracy of FEP depends critically on the quality of the starting protein-ligand structure.

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. 137, 2695-2703 (2015).
2. Cournia, Z. et al. "Best Practices for Quantitative Binding Free Energy Calculations." J. Chem. Inf. Model. (2023).
3. Schrödinger. "FEP+ Benchmark Results." Technical report (2023).
4. Song, L.F. & Merz, K.M. "Evolution of Alchemical Methods in Drug Discovery." J. Chem. Inf. Model. (2020).
5. Hahn, D.F. et al. "Best Practices for Constructing, Preparing, and Evaluating Protein-Ligand Binding Affinity Calculations." Living J. Comput. Mol. Sci. (2023).