Enhanced Sampling

Most enhanced sampling methods rely on collective variables (CVs) —low-dimensional descriptors of the system's slow degrees of freedom. The bias potential is applied along these predefined coordinates.

• Distance CVs: Interatomic distances, e.g., ligand-receptor separation
• Angular CVs: Dihedral angles governing torsional transitions
• Coordination numbers: Count of contacts between groups of atoms
• Path CVs: Progress along a predefined reaction pathway

The choice of CV critically determines sampling efficiency. Poorly chosen CVs can lead to hysteresis and incomplete exploration of the free energy landscape. Advanced methods like Time-lagged Independent Component Analysis (TICA) help identify optimal CVs from simulation data.

Methods like metadynamics construct a bias potential on-the-fly by periodically depositing Gaussian functions along the CV space. This discourages the system from revisiting already-explored regions.

• Well-Tempered Metadynamics: Scales Gaussian heights down as bias accumulates, ensuring asymptotic convergence to the true free energy surface
• Multiple Walker Metadynamics: Parallel simulations share a common bias potential, dramatically accelerating exploration
• Bias Exchange Metadynamics: Multiple replicas bias different CVs and periodically exchange configurations

The accumulated bias provides a direct estimate of the underlying free energy surface: F(s) = -V_bias(s) at convergence.

Replica Exchange Molecular Dynamics (REMD) runs multiple non-interacting replicas at different temperatures or Hamiltonians, periodically attempting to swap configurations between adjacent replicas.

• Temperature REMD: Replicas span a temperature ladder; high-temperature replicas cross barriers easily, and successful swaps propagate low-energy configurations downward
• Hamiltonian REMD: Replicas differ in potential energy scaling, e.g., selectively scaling solute-solvent interactions to enhance sampling of specific degrees of freedom
• Replica Exchange with Solute Tempering (REST2): Scales only the solute potential, requiring far fewer replicas than standard T-REMD

Exchange acceptance follows the Metropolis criterion, ensuring detailed balance is maintained across the entire replica ensemble.

Umbrella sampling imposes harmonic restraints to sample overlapping windows along a reaction coordinate, then unbias the distributions to reconstruct the Potential of Mean Force (PMF) .

• Harmonic restraint: V = k(ξ - ξ₀)² applied at each window center ξ₀
• Overlap requirement: Adjacent windows must have sufficient histogram overlap for reliable unbiasing
• WHAM: The Weighted Histogram Analysis Method iteratively solves self-consistent equations to combine biased distributions into the unbiased PMF
• MBAR: The Multistate Bennett Acceptance Ratio provides a statistically optimal generalization of WHAM with lower variance

This method is particularly effective for calculating free energy profiles of ligand binding and ion permeation through channels.

GaMD smoothens the potential energy surface without requiring predefined collective variables, making it suitable for systems where reaction coordinates are unknown.

• Dihedral boost: Adds a harmonic boost potential to dihedral angles to accelerate torsional transitions
• Dual boost: Applies both dihedral and total potential boosts simultaneously
• Threshold energy: Boost is applied only when the system potential falls below a threshold, preserving the shape of the original energy surface
• Reweighting: Cumulant expansion to the second order recovers the unbiased canonical ensemble from the biased simulation

GaMD has been successfully applied to protein folding, ligand binding, and GPCR activation studies without prior knowledge of the transition pathway.

The ultimate goal of enhanced sampling is to reconstruct the free energy landscape —a map of the system's thermodynamic stability as a function of collective variables.

• Free energy surface: F(s) = -kBT ln P(s), where P(s) is the unbiased probability distribution along CV s
• Minimum free energy path: The most probable transition pathway connecting metastable states
• Kinetic rates: Transition rates between basins can be estimated from barrier heights using Kramers' rate theory or transition path sampling
• Markov State Models: Combine many short biased simulations into a kinetic network model describing long-timescale dynamics

Accurate landscape reconstruction enables quantitative prediction of binding affinities, conformational populations, and transition mechanisms.

A history-dependent enhanced sampling method that accelerates rare events by discouraging revisitation of previously explored states. A Gaussian bias potential is deposited along predefined collective variables (CVs) at regular intervals, progressively filling free energy minima until the system can freely diffuse across barriers. Well-tempered metadynamics converges the bias to an exact estimate of the free energy surface by scaling Gaussian heights as exploration proceeds.

A parallel tempering technique that runs multiple non-interacting replicas of the system at different temperatures or Hamiltonian parameters simultaneously. At periodic intervals, the algorithm attempts to exchange configurations between neighboring replicas based on a Metropolis criterion. High-temperature replicas easily cross energy barriers, and successful swaps propagate these rare configurations down to the target temperature, dramatically improving conformational sampling without biasing forces.

A method for calculating the potential of mean force (PMF) along a reaction coordinate by dividing the path into overlapping windows. Each window applies a harmonic restraint to confine sampling to a narrow region. The biased distributions are subsequently unbiased and combined using the Weighted Histogram Analysis Method (WHAM) or Multistate Bennett Acceptance Ratio (MBAR) to reconstruct the full free energy profile.

A CV-free enhanced sampling method that smoothens the potential energy surface by adding a harmonic boost potential to dihedral angles and a non-harmonic boost to the total potential. By raising the energy of local minima more than barriers, GaMD accelerates transitions between low-energy states without requiring predefined collective variables. The boost potential can be reweighted to recover the original free energy landscape.

A kinetic network framework that discretizes phase space into metastable states and estimates a transition probability matrix from many short, parallel simulations. MSMs bridge the gap between accessible simulation timescales and experimentally relevant biological timescales by propagating the Markov chain. Time-lagged independent component analysis (tICA) identifies the slowest degrees of freedom for optimal state decomposition.

A deep generative modeling approach using normalizing flows to learn an invertible mapping between a simple latent distribution and the complex Boltzmann distribution of a molecular system. Once trained, Boltzmann generators can sample statistically independent configurations in a single forward pass and compute exact free energy differences, bypassing the need for sequential MD integration entirely.