[Paper Review] Enhanced sampling methods for molecular dynamics simulations

Jérôme Hénin, Tony Lelièvre+3 more|arXiv (Cornell University)|Feb 8, 2022

Protein Structure and Dynamics24 citations

TL;DR

This LiveCoMS perpetual review surveys enhanced sampling methods for molecular dynamics, unifying their principles, classifications, and relationships, to help researchers choose appropriate approaches and understand their connections.

ABSTRACT

Enhanced sampling algorithms have emerged as powerful methods to extend the utility of molecular dynamics simulations and allow the sampling of larger portions of the configuration space of complex systems in a given amount of simulation time. This review aims to present the unifying principles and differences of many of the computational methods currenly used for enhanced sampling in molecular simulations of biomolecules, soft matter and molecular crystals. Indeed, despite the apparent abundance and divergence of such methods, the principles at their core can be boiled down to a relatively limited number of statistical and physical principles. To enable comparisons, the various methods are introduced using similar terminology and notation. We then illustrate in which ways many different methods combine principles from a smaller class of enhanced sampling concepts. This review is intended for scientists with an understanding of the basics of molecular dynamics simulations and statistical physics who want a deeper understanding of the ideas that underlie various enhanced sampling methods and the relationships between them. This living review is intended to be updated to continue to reflect the wealth of sampling methods as they continue to emerge in the literature.

Motivation & Objective

• Clarify the scope and purpose of enhanced sampling in MD for biomolecules, soft matter, and crystals.
• Unify underlying statistical and physical principles behind diverse methods.
• Classify and compare how different enhanced sampling schemes operate and relate to each other.
• Provide guidance on selecting methods and parameters, and summarize available software implementations.
• Highlight the trade-offs between exploration and accurate equilibrium sampling and discuss kinetic information considerations.

Proposed method

• Present a unified terminology and notation to enable cross-method comparisons.
• Classify methods along core principles such as partitioning vs overlapping ensembles and equilibrium vs non-equilibrium goals.
• Describe major families of enhanced sampling methods (e.g., metadynamics, ABF, replica exchange, expanded ensembles, adaptive seeding, selective acceleration) and their core mechanisms.
• Explain free energy estimators (TI, BAR/MBAR, WHAM) and how they apply across methods.
• Discuss diagnostics, convergence, and practical considerations for method choice and parameter tuning.
• Summarize software implementations and practical guidelines for applying these methods.

Experimental results

Research questions

• RQ1 What are the fundamental statistical and physical principles that underpin enhanced sampling methods in MD?
• RQ2 How do different enhanced sampling schemes relate to each other within a unified framework?
• RQ3 How should researchers choose among methods and parameters for a given system and goal (thermodynamic vs kinetic information, accuracy vs exploration)?

Key findings

• The review provides a unified framework to understand diverse enhanced sampling methods.
• It identifies eight methodological classes and highlights partitioning versus overlapping ensemble structures.
• It shows how common free energy estimators (TI, BAR/MBAR, WHAM) apply across methods.
• It discusses when kinetics can be preserved or recovered and when methods primarily target equilibrium ensembles.
• It emphasizes the living nature of the review and invites community contributions via GitHub.
• It catalogs public software implementations relevant to these methods.