Introduction to Performing Metadynamics Simulations with Desmond

Tutorial Created with Software Release: 2026-2

Topics: Biologics Drug Discovery, Small Molecule Drug Discovery, Structure Prediction & Target Enablement

Products Used: Desmond, Force Field Bundle,

Tutorial Content

1. Introduction to Metadynamics Simulations
2. Importing and Preparing Structures
3. Metadynamics for Conformational Sampling
4. Obtaining a potential energy surface for nucleotide unstacking with well-tempered metadynamics
5. Conclusion and References

1. Introduction to Metadynamics Simulations

A key limitation of molecular dynamics (MD) simulations is that getting the simulated system out of a local energy minimum to explore more of its conformational space requires extensive (and in the case of large systems, prohibitively costly) sampling over long simulation times. Metadynamics simulations enable a greater sampling of the conformational space of a molecule in a shorter amount of computer time compared to unbiased molecular dynamics.

In molecular dynamics simulations, the total force is calculated from a potential energy function built up of simplified representations of the forces between the atoms (i.e. the force field terms), and then Newton’s equations of motions are used to push the atoms through time over a specified change in time. Metadynamics simulations artificially alter the potential energy function to discourage the system from remaining in conformations which were already explored during the simulation. This is achieved by adding artificial inverse Gaussian functions to the potential energy, which increase the total potential energy of the molecule. As the overlapping Gaussian potentials accumulate during the course of the simulation, they allow the molecule to overcome barriers between local minima and explore the energy landscape. To make these simulations feasible, you need to choose collective variables (CVs) which describe the degree(s) of freedom you wish to bias your system towards exploring.

Metadynamics simulations are therefore useful tools for generating diverse conformers or accurately mapping out free energy landscapes. In this tutorial, you will explore several applications of metadynamics to simple example systems. You will perform a simulation of the small-molecule drug zopolrestat in water for generating diverse conformers. In a second example, you will perform well-tempered metadynamics with two collective variables to explore the stacking-unstacking equilibrium in the ApA dinucleotide.

2. Importing and Preparing Structures

Download and extract the zip file here to your working directory to obtain the inputs and reference outputs for this tutorial: schrodinger.com/sites/default/files/s3/release/current/Tutorials/zip/metadynamics.zip

The input structure provided for zopolrestat has already been prepared using LigPrep and the System Builder was used to place it into a simulation box with water and one sodium counterion.

To learn how to perform these steps yourself, you can follow the instructions below with the zopolrestat.sdf structure provided in the tutorial zip archive or your own example system.

The input structure provided for the dinucleotide system was built using Build Biopolymer from Sequence, and prepared for simulation using the System Builder. As usual when simulating nucleic acids, we have added sodium and chloride ions equivalent to a 0.15M salt concentration in addition to neutralizing the system.

3. Metadynamics for Conformational Sampling

In this section, you will use metadynamics to generate conformers for a small molecule drug (zopolrestat, an aldose reductase inhibitor currently in phase II trials). While commonly used conformer generation models such as MacroModel and ConfGen use torsional sampling and rudimentary energetics to quickly return reasonably low-energy conformers, using MD-based sampling has the advantage of an explicit solvent model and more natural sampling. Unbiased MD can however get stuck in local minima or only sample very low-energy conformations, potentially missing the bioactive conformation completely.

Here, metadynamics can be helpful to speed up and extend the sampling. This type of metadynamics simulation is also quite straightforward to set up and analyze, as the CV definition is simple and there’s no need to carefully converge the PES to obtain accurate energetics if the goal is to obtain a diverse set of conformations.

3.1 Setting up a standard metadynamics simulation for a small molecule in water

In this simple example, you will use a distance CV between two atoms in the zopolrestat molecule to sample multiple torsional degrees of freedom at once. This approach is not suited to obtain a converged and accurate PES, but can efficiently explore the conformational space available to the molecule in solution.

3.2 Extracting conformations from the metadynamics trajectory

You can now use the usual tools for trajectory analysis to view, extract, or cluster the conformers sampled during the simulation. For a general introduction on how to interact with MD trajectories in Maestro, see the Introduction to MD Trajectory Analysis with Desmond tutorial.

Here, you will briefly check the trajectory and visualize the sampling for the relevant torsional degrees of freedom. Then, you will extract the ligand structures from the trajectory and cluster them to identify representative ligand conformations for further use.

Sidenote: Clustering tools in the Schrödinger suite

There are several panels in the Schrödinger suite with which you can perform clustering. They perform similar calculations in the backend, but provide different options for featurization and clustering methods. Feel free to explore the panels and their documentation pages to find which one is best suited to your needs:

• Conformer Cluster Panel – straightforward to configure for geometry-based clustering of small molecules.
• Canvas Similarity and Clustering Panel – provides GUI access to most types of featurizations and clustering methods.
• Trajectory Frame Clustering Panel
• Spectral Clustering Panel
• Clustering Based on Volume Overlap Panel

4. Obtaining a potential energy surface for nucleotide unstacking with well-tempered metadynamics

In this section, we will explore the energetics of the stacking-unstacking equilibrium in a rudimentary dinucleotide system. We’ll set up the simulation parameters to allow accurate mapping of the potential energy surface. After the simulation, we will confirm the convergence and view the resulting energy landscape.

4.1 Setting up a well-tempered metadynamics simulation with two collective variables

For this system, two collective variables will be needed to obtain meaningful and interpretable results: the center-of-mass distance between the nucleobases, and the torsion around the glycosidic bond to capture their relative angles. In order to obtain a well-converged potential energy surface for this system, we will enable well-tempered metadynamics by providing the biasing parameter kTemp.

Sidenote: Building custom CVs, restraints, and metadynamics methods

Do you want to define your own custom collective variables beyond those supported in the .msj file, use custom restraint potentials, or implement complex simulation workflows (e.g. bias exchange or parallel bias metadynamics)?

You can use the enhanced sampling plugin integrated with Desmond to take full control over your metadynamics simulations. Learn more in the Custom Enhanced Sampling documentation.

4.2 Analyzing the convergence and results of the metadynamics simulation

Due to the flexible nature of the metadynamics approach, the analysis and interpretation steps will usually be tailored to the specific scientific questions of the project. Here, we will focus on confirming that the simulation has converged enough for its results to be meaningfully interpretable, and highlight how to visualize the resulting potential energy surface.

More information on the metadynamics_convergence_analysis.py script

In a nutshell, this script splits the productive stage of the simulation into chunks and compares various properties between consecutive chunks. The smaller the differences between consecutive chunks, the more converged we consider the simulation.

The arguments we passed to the script were, in order:

• The output configuration file (<jobname>-out.cfg), containing the metadynamics settings.
• The kernel sequence file (<jobname>.kerseq), logging all biasing potentials which were applied during the simulation.
• The output file name for the report.
• The assignment of the analysis chunks. We specified this as the number of -bins into which to split the simulation equidistantly. Keep the size of the chunks large enough to allow for decent sampling within each chunk and reduce statistical noise — approximately 10 ns per chunk is a good size to aim for.

This script is limited to simulations with one or two collective variables defined in the .msj file, with or without well-tempering. It also works on simulations which have crashed rather than finished cleanly. Simulations with custom .pot files are not supported.

5. Conclusion and References

In this tutorial, you learned how to set up and analyze metadynamics simulations for conformer exploration and obtaining accurate potential energy surfaces with well-tempered metadynamics.