What is OpenMM?

OpenMM is a high-performance molecular dynamics simulation toolkit that enables GPU-accelerated simulations of biomolecular systems. Developed by the Pande and Chodera labs at Stanford and MSKCC, OpenMM provides a flexible platform for studying protein dynamics, conformational changes, and ligand-bound complexes with industry-standard force fields like AMBER, CHARMM, and OpenFF small-molecule parameters.

Molecular dynamics (MD) simulations model atomic motion over time by numerically integrating Newton's equations of motion. Each atom experiences forces from bonded interactions (bonds, angles, dihedrals) and non-bonded interactions (electrostatics, van der Waals). By tracking these forces over femtosecond timesteps, MD reveals how proteins move, fold, and interact with their environment on nanosecond to microsecond timescales.

ProteinIQ runs OpenMM 8.5.1 with classical AMBER14-SB or CHARMM36 protein force fields. It can also load a complete new-style AMBER topology and matching ASCII or NetCDF restart through OpenMM's native AmberPrmtopFile and AmberInpcrdFile readers. Separate tools such as Orb v3 support machine-learning potentials; this OpenMM configuration does not offer them. Simulations run on GPU hardware, and throughput depends strongly on atom count, solvent model, constraints, and platform availability.

How does molecular dynamics work?

The physics of motion

At each timestep, OpenMM calculates the potential energy of the system and derives forces from the gradient:

$F_i = -\nabla_i U(r_1, r_2, \ldots, r_N)$

where $U$ is the potential energy function and $r_i$ are atomic coordinates. Forces accelerate atoms according to Newton's second law ($F = ma$), and velocities are integrated to update positions.

The potential energy function (force field) decomposes into bonded and non-bonded terms:

$U = U_{bonds} + U_{angles} + U_{dihedrals} + U_{electrostatics} + U_{vdW}$

Time integration

OpenMM uses Langevin dynamics, which adds friction and random forces to simulate coupling with an implicit heat bath:

$m_i \frac{d^2 r_i}{dt^2} = F_i - \gamma m_i \frac{dr_i}{dt} + \sqrt{2 m_i \gamma k_B T} \cdot R(t)$

The friction coefficient $\gamma$ and temperature $T$ control thermal equilibration. Random forces $R(t)$ sample from a Gaussian distribution to maintain the target temperature.

Force fields

AMBER14-SB is the default force field, offering well-validated parameters for proteins based on extensive benchmarking against NMR data and quantum calculations. It uses fixed partial charges and Lennard-Jones potentials for non-bonded interactions.

CHARMM36 provides an alternative with different charge assignments and dihedral parameterization. It offers superior lipid parameters if you plan membrane simulations or protein-lipid interaction studies.

Both force fields assume a classical treatment of electron density—atoms are point masses with fixed charges. This approximation works well for most protein simulations but breaks down for processes involving electronic polarization, charge transfer, or bond breaking.

Solvation models

Explicit water surrounds the protein with thousands of discrete water molecules. AMBER14-SB runs use the TIP3P-FB water parameters; CHARMM36 runs use the water model supplied with CHARMM36. Explicit solvent captures hydrogen bonding, hydrophobic effects, and solvent-mediated interactions with high fidelity.

The computational cost scales with system size. A 100-residue protein in a typical water box requires ~30,000 total atoms, with water molecules comprising ~90% of the atoms.

Implicit solvent (GBn2) replaces discrete water molecules with a continuum dielectric. The generalized Born model approximates electrostatic screening by surrounding atoms with a polarizable medium characterized by dielectric constant and Born radii.

Implicit solvent reduces system size dramatically (protein atoms only) and eliminates water equilibration time. However, it sacrifices detailed hydrogen bonding and can distort conformational preferences for surface-exposed residues.

Input parameters

Starting point

Prepare a structure is the default workflow. Submit a PDB or mmCIF protein structure and, optionally, one positioned SDF or MOL2 ligand. ProteinIQ repairs the structure, assigns the selected force field, adds hydrogens, and configures solvent before simulation.

Start from AMBER files accepts exactly two matching files:

• A new-style AMBER topology (.prmtop or .parm7) containing the complete system and all force-field parameters.
• An AMBER ASCII coordinate/restart file (.inpcrd, .rst, or .rst7) or NetCDF restart (.ncrst). Coordinates are required; velocities and periodic box vectors are loaded automatically when present.

AMBER mode passes these files directly to OpenMM. It does not run PDBFixer, add or remove atoms, reparameterize a ligand, add solvent or ions, assign protonation states, minimize, or equilibrate the submitted state. Any protein, ligand, cofactor, ion, water, or nucleic acid needed by the simulation must already be represented consistently in both files.

Choose Explicit only for a prepared periodic AMBER system with box vectors. OpenMM uses PME and, by default, an NPT Monte Carlo barostat; turn the barostat off for NVT. Choose Implicit GBn2 only when the topology was prepared with compatible AMBER intrinsic radii. In implicit mode, the salt concentration defaults to OpenMM's native 0 M value.

An AMBER restart transfers public state—positions, available velocities, and available box vectors—but is not an exact continuation of AMBER's engine-internal state. Thermostat and barostat internals, random-number state, step count, and simulation time are not portable through this format, so the returned production log starts at 0 ps. If the coordinate file has no velocities, OpenMM initializes them at the requested temperature.

Simulation settings

Simulation duration controls how long the production simulation runs. Longer simulations sample more conformational space and take more time.

• 1 ns (quick test) — Verify the system is stable and check for obvious problems
• 10 ns (equilibration) — Standard equilibration check; sufficient to detect major issues
• 50 ns (standard) — Production-quality sampling for small conformational changes
• 100 ns (production) — Extended sampling for larger motions or binding events

Solvation model determines how solvent is treated. In structure-preparation mode, explicit water is added with TIP3P-FB for AMBER14 or the packaged CHARMM36 water model. In AMBER-file mode, explicit solvent is never added or changed: the submitted periodic system is used as-is. Implicit solvent (GBn2) uses OpenMM's native GBn2 system construction and is typically slower on this GPU configuration.

Force field selects the energy function parameters. AMBER14-SB is recommended for most protein work. Choose CHARMM36 if planning membrane simulations or if your workflow uses CHARMM-derived parameters downstream.

Ligand file is optional and supplies both chemistry and positioned coordinates for one organic small molecule. Upload a 3D SDF or MOL2 pose in the same coordinate frame as the protein, or upload a matching ligand file for coordinates already embedded in the protein structure. OpenMM adds a separate submitted pose to the system; it does not dock or reposition it. Metal complexes and ligands requiring custom parameters are not supported.

Ligand coordinate source defaults to the submitted SDF/MOL2 pose. Choose Match ligand embedded in protein only when the PDB/mmCIF already contains the same ligand coordinates and the ligand file supplies matching chemistry.

Unparameterized heterogens applies whenever the submitted structure is prepared without a verified embedded ligand. By default, the run stops and reports non-water components that PDBFixer would remove—even when a separate positioned ligand is supplied. Choose explicit removal only when excluding those ligands, cofactors, or ions is scientifically intended.

Input structures are limited to 100,000 atoms and ligands to 300 heavy atoms. Before dynamics begins, OpenMM also checks the prepared atom count, requested duration, and save interval; requests estimated to exceed 1 GiB of trajectory coordinates must use a longer save interval, shorter duration, or smaller system.

Environment

Temperature sets the simulation temperature in Kelvin. 300 K (27°C) is standard physiological temperature. Use higher temperatures (350-400 K) for enhanced sampling of slow processes, or lower temperatures for cold-adapted proteins.

Pressure controls the barostat target in bar for explicit-solvent NPT simulations. 1.0 bar is standard atmospheric pressure. Implicit-solvent simulations use NVT and do not apply pressure.

Ionic strength specifies NaCl concentration in molar for explicit solvent. 0.15 M matches physiological conditions. The system is neutralized with counterions and excess salt is added to reach the target concentration. It does not apply to implicit-solvent runs.

pH determines protonation states of titratable residues (histidine, aspartate, glutamate, lysine, arginine). Standard protonation at pH 7.0 is appropriate for most cytoplasmic proteins. Adjust for proteins in acidic compartments (lysosomes, pH 4.5-5.0) or extracellular environments.

Output options

Save interval controls trajectory frame frequency. 50 ps produces manageable file sizes while capturing relevant dynamics. Use 10 ps for detailed analysis of fast motions or transition states. Use 100 ps for long simulations where disk space is limited.

Remove water from output strips only water from the aligned viewing trajectory while retaining ligands, cofactors, ions, proteins, and nucleic acids. The complete engine-native trajectory and matching full-system final structure are returned separately.

Advanced settings

Timestep is the integration step size. 2 fs is standard when hydrogen bonds are constrained. 4 fs (HMR) uses hydrogen mass repartitioning—hydrogen masses are increased while heavy atom masses decrease to maintain total mass. This allows larger timesteps without instability but requires validation for your specific system.

Minimization steps controls energy minimization before dynamics. Minimization removes bad contacts and steric clashes from the initial structure. 1000 steps suffice for most structures. Increase for structures with severe clashes or added loops.

Equilibration time sets the dynamics phase before production data collection. Explicit-solvent runs equilibrate with the NPT barostat active; implicit-solvent runs equilibrate at NVT. 0.5 ns is the default.

Bond constraints determines which bonds are held rigid:

• H-bonds only — Constrain bonds involving hydrogen, enabling 2 fs timesteps
• All bonds — Constrain all bonds

The 4 fs option applies 4 amu hydrogen mass repartitioning and requires bond constraints.

Understanding the results

Trajectory output

OpenMM produces a trajectory file containing atomic coordinates at each save interval. The trajectory can be visualized in molecular viewers (VMD, PyMOL, Chimera) to observe protein motion over time.

Key observations from trajectory analysis:

• RMSD — Root-mean-square deviation from the starting structure indicates overall conformational stability
• RMSF — Root-mean-square fluctuation per residue reveals flexible regions
• Radius of gyration — Measures protein compactness; changes indicate unfolding or large conformational shifts
• Secondary structure — Track helix/sheet content over time to detect folding/unfolding events

These built-in stability analyses use the protein atoms only, independently of whether the downloadable processed trajectory retains explicit water.

Energy monitoring

The native state_data.csv log reports step, time, potential energy, kinetic energy, total energy, and temperature at regular intervals. Stable simulations show:

• Potential energy fluctuating around a constant value (not drifting)
• Temperature fluctuating around the target value (300 K ± 5 K typical)
• No sudden energy spikes indicating steric clashes or numerical instabilities

Common issues

Structure explodes — Usually indicates bad starting contacts or missing parameters. Run energy minimization first with PDB Fixer. Ensure all atoms have appropriate force field parameters.

Temperature drifts — Check that the thermostat is functioning correctly. Very large systems may need longer equilibration.

Protein unfolds — May indicate force field issues with your specific protein or unrealistic starting conditions. Verify the structure quality with MolProbity before simulation.

When to use OpenMM

Use molecular dynamics when

• Studying protein flexibility and conformational changes
• Investigating binding site dynamics and ligand accommodation
• Relaxing predicted structures from AlphaFold2 or ESMFold
• Generating conformational ensembles for downstream analysis
• Validating protein stability under different conditions

Consider alternatives when

• You only need a single static structure — use structure prediction
• Studying chemical reactions or bond breaking — MD cannot model these
• Simulating very large systems (viruses, ribosomes) — requires specialized approaches
• Timescales exceed microseconds — enhanced sampling methods may be needed

Workflow integration

Structure preparation

Before running MD, ensure your structure is simulation-ready:

1. Fix structural issues — Use PDB Fixer to add missing atoms, residues, and hydrogens
2. Validate geometry — Run MolProbity to check for clashes and outliers
3. Assign protonation — Use PROPKA to determine appropriate pH if unsure

Post-simulation analysis

After MD completes:

1. Visualize trajectory — Use PDB Viewer to inspect representative frames
2. Calculate RMSD — Use RMSD Calculator to compare conformations
3. Analyze secondary structure — Use DSSP on trajectory frames
4. Check compactness — Use Radius of Gyration calculator

Docking workflows

MD simulations can prepare receptors for docking or validate docking poses:

1. Run brief MD to generate receptor conformational ensemble
2. Extract representative snapshots for ensemble docking
3. Use AutoDock Vina, Gnina, or DiffDock for docking
4. Run MD on docking complexes to assess pose stability

Frequently asked questions

Is OpenMM free to use?

OpenMM itself is open-source software. ProteinIQ OpenMM jobs use paid compute credits because they run on dedicated GPU infrastructure; guest and free plans cannot currently submit this tool.

How long do simulations take?

Simulation time depends on system size, solvation model, duration, and platform. GBn2 implicit solvent can be substantially slower than optimized explicit-solvent GPU kernels despite using fewer atoms. Jobs have a 12-hour runtime limit and may return completed partial artifacts if execution ends earlier.

What's the difference between explicit and implicit solvent?

Explicit solvent adds discrete water molecules that interact with the modeled system. Implicit solvent replaces water with a continuum approximation and sacrifices detailed solvent interactions. On this GPU configuration, GBn2 implicit solvent is not a speed shortcut; use it only when its physical approximation matches the question.

Should I use AMBER or CHARMM force fields?

For standard protein simulations, AMBER14-SB is recommended and well-validated. CHARMM36 offers advantages for membrane simulations due to superior lipid parameters. Both produce reasonable results for most proteins—the differences are typically smaller than uncertainties from other sources (sampling, starting structure quality).

How do I know if my simulation is stable?

Check these indicators:

• Energy stays within normal fluctuation range (no spikes or drift)
• Temperature fluctuates around target value
• RMSD plateaus rather than continuously increasing
• Visual inspection shows no unfolding or unrealistic motions

Can I simulate protein-ligand complexes?

Yes. Submit a protein structure and a coordinate-bearing 3D SDF or MOL2 pose in the same coordinate frame, or provide a matching ligand file for an embedded ligand. The ligand is parameterized with OpenFF 2.3.0 and added at its submitted coordinates when it is separate. This is not a docking step, and metal-containing ligands or coordination complexes require custom force-field parameters outside this tool.

What should I do if the simulation crashes?

Most crashes result from bad starting structures. Try:

1. Run thorough energy minimization with more steps
2. Fix structural issues with PDB Fixer
3. Check for clashing atoms with MolProbity
4. Ensure all atoms have valid force field parameters
5. For ligand complexes, confirm that the SDF/MOL2 contains a valid 3D pose in the protein coordinate frame

How much conformational sampling do I need?

This depends on your question:

• Stability check: 1-10 ns suffices to detect major problems
• Local flexibility: 10-50 ns captures side chain rotations and loop motions
• Domain motions: 50-100 ns may capture larger conformational changes
• Rare events: Enhanced sampling methods may be needed for processes slower than 100 ns

Can I continue a simulation from where it left off?

You can start production dynamics from an AMBER .prmtop plus matching .inpcrd, .rst, .rst7, or .ncrst. OpenMM imports positions, available velocities, and available box vectors without rerunning structure preparation, minimization, or equilibration. This is a physical-state transfer rather than an exact AMBER engine continuation: internal thermostat, barostat, random-number, time, and step state are not contained in an OpenMM checkpoint and cannot be restored from an AMBER restart.

References

Eastman, P., Galvelis, R., et al. (2024). OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials. J. Phys. Chem. B, 128(1), 109-116. DOI: 10.1021/acs.jpcb.3c06662

Maier, J.A., et al. (2015). ff14SB: Improving the Accuracy of Protein Side Chain and Backbone Parameters from ff99SB. J. Chem. Theory Comput., 11(8), 3696-3713.

Huang, J., et al. (2017). CHARMM36m: an improved force field for folded and intrinsically disordered proteins. Nat. Methods, 14, 71-73.