Absolute Binding Free Energy

A computational technique that calculates the free energy difference between two states by simulating a non-physical thermodynamic pathway. The ligand is gradually decoupled from its environment through a series of intermediate states (λ-windows), where interactions are scaled from fully present to fully absent.

• Key distinction: ABFE is a specific application of alchemical methods where the ligand is completely annihilated in both the bound and unbound states.
• Core advantage: Avoids the need to physically pull the ligand along a dissociation path, which can be hindered by high energy barriers.
• Common implementations: Uses thermodynamic integration (TI) or free energy perturbation (FEP) to accumulate the work of decoupling.

The standard thermodynamic cycle for computing absolute binding free energy. It decomposes the binding process into two independent alchemical legs:

• Leg 1 (Bound): The ligand is decoupled from the solvated protein-ligand complex, removing its interactions with both the protein and solvent while restrained in the binding pocket.
• Leg 2 (Unbound): The ligand is decoupled from pure solvent, removing only its solvent interactions.
• Restraint correction: A separate analytical or numerical correction accounts for the loss of translational, rotational, and conformational freedom upon binding.
• Cycle closure: ΔG_bind = ΔG_restraint + ΔG_decouple(bound) − ΔG_decouple(unbound)

The free energy profile along a chosen reaction coordinate (e.g., the center-of-mass distance between ligand and binding pocket). Unlike alchemical methods, PMF-based ABFE physically pulls the ligand out of the binding site.

• Umbrella Sampling: Applies harmonic restraints at discrete positions along the path to ensure adequate sampling of high-energy regions.
• WHAM (Weighted Histogram Analysis Method): Unbiases the overlapping window distributions to reconstruct the continuous free energy profile.
• Challenge: Requires identifying a slow, barrier-free coordinate; orthogonal degrees of freedom may introduce hysteresis if not adequately sampled.
• Comparison: PMF methods are more intuitive but often suffer from convergence issues compared to alchemical approaches for complex binding pockets.

A statistically optimal estimator for calculating free energy differences by simultaneously analyzing data from all intermediate alchemical states, rather than pairwise adjacent windows.

• MBAR formulation: Solves a set of self-consistent equations to minimize the statistical variance of the free energy estimate across the entire thermodynamic path.
• Superior to BAR: While the Bennett Acceptance Ratio (BAR) uses only two adjacent states, MBAR incorporates all states, reducing uncertainty by up to 30%.
• Overlap requirement: Requires sufficient phase-space overlap between each state and every other state; poor overlap leads to large uncertainties.
• Implementation: Available in pymbar, alchemlyb, and integrated into major simulation packages like GROMACS and OpenMM.

When a ligand is decoupled inside a binding pocket, artificial restraints (harmonic, flat-bottom, or Boresch-style) must be applied to prevent it from drifting away as interactions vanish. The free energy contribution of these restraints must be analytically or numerically subtracted.

• Boresch restraints: Six harmonic potentials (one distance, two angles, three dihedrals) that define the ligand's relative orientation to the protein.
• Analytical correction: Derived from the partition function of a rigid rotor and harmonic oscillator, requiring no additional simulation.
• Numerical correction: Computed via a separate alchemical simulation that releases the restraints in the fully decoupled state.
• Standard state: The correction also accounts for the conversion from the restrained bound volume to the 1 M standard state (1 molecule / 1660 ų).

Converging ABFE calculations requires exhaustive sampling of ligand and protein conformations. Enhanced sampling techniques accelerate this exploration:

• Replica Exchange with Solute Tempering (REST2): Scales the ligand-protein and ligand-solvent interactions in different replicas, allowing the ligand to escape local minima while the solvent remains unperturbed.
• Metadynamics: Deposits a history-dependent bias potential along collective variables to discourage revisiting sampled conformations, useful for exploring slow protein side-chain rearrangements.
• Hamiltonian Replica Exchange: Exchanges configurations between adjacent λ-windows, improving mixing and reducing the number of independent simulations required.
• Alchemical Metadynamics: Combines alchemical λ-dynamics with metadynamics to simultaneously sample both the chemical and conformational spaces.