How to Choose a Force Field for MD Simulation: AMBER vs CHARMM vs GROMOS vs OPLS

The AMBER protein force fields — particularly ff14SB and its successor ff19SB — are the most widely cited force fields in the drug discovery and structural biology literature. They were developed at UC San Francisco and have been extensively validated against NMR chemical shifts, J-couplings, and experimental secondary structure populations across a diverse range of proteins.

ff19SB is the current recommended version for protein simulations. It was trained using a significantly larger quantum mechanical dataset than ff14SB and shows improved backbone and side-chain conformational populations. The practical difference between them is often small for stable, well-folded proteins, but becomes meaningful for intrinsically disordered proteins and peptides.

AMBER’s nucleic acid force fields — RNA.OL3 for RNA and DNA.OL15 for DNA — are the most extensively validated in the field and are the default choice for nucleic acid simulations. For small molecule parameterization, GAFF (General AMBER Force Field) and its successor GAFF2 provide parameters for arbitrary organic molecules compatible with AMBER protein force fields.

CHARMM36 and its variant CHARMM36m (optimized for intrinsically disordered proteins and protein-lipid systems) are the dominant force fields for membrane protein simulation. This dominance stems from two factors: CHARMM’s comprehensive lipid parameter set — covering virtually every biologically relevant lipid including PC, PE, PS, PI, cholesterol, and their variants — and the CHARMM-GUI web server, which provides automated system building for membrane protein simulations using CHARMM parameters natively.

CHARMM36m specifically improved the treatment of disordered regions and flexible loops compared to CHARMM36, while maintaining compatibility with the same lipid parameters. For any project involving a membrane protein, CHARMM36m is the current recommendation.

For small molecule parameterization within CHARMM workflows, CGenFF (CHARMM General Force Field) provides parameters for drug-like molecules. CGenFF parameters can be generated automatically through the ParamChem web server, with penalty scores indicating how well-parameterized each molecule is.

GROMOS takes a fundamentally different approach from AMBER and CHARMM: it uses a united-atom representation where aliphatic hydrogen atoms are not represented explicitly but are instead merged into the heavy atom they’re bonded to. This reduces the number of atoms in the simulation by roughly 30–40%, making GROMOS simulations faster for equivalent system sizes.

The tradeoff is reduced accuracy for properties that depend on explicit hydrogen positions — particularly hydrogen bonding geometry and hydrophobic interactions. For many applications this is acceptable; for others (NMR chemical shift prediction, explicit hydrogen bonding analysis) it’s a meaningful limitation.

GROMOS was developed at the University of Groningen and ETH Zurich with particular attention to reproducing experimental free energies of solvation and transfer between aqueous and nonpolar environments. The current recommended version is 54A7. It’s less commonly used for new projects than AMBER or CHARMM, but remains in use in European research groups and for free energy calculation workflows where its thermodynamic accuracy is valued.

OPLS-AA was developed at Yale by Bill Jorgensen’s group with a specific focus on reproducing experimental thermodynamic properties of organic liquids and solvation free energies. Its parameters were fitted to condensed-phase experimental data — densities, heats of vaporization, free energies of hydration — rather than gas-phase quantum mechanical calculations, giving it distinctive strengths for small organic molecule simulation.

For drug-like molecules and organic solvent simulations, OPLS-AA and its commercial extension OPLS3e/OPLS4 (from Schrödinger) are highly competitive with GAFF and CGenFF. OPLS4 in particular shows excellent performance in recent benchmarks for protein-ligand binding free energy calculations and is the force field underlying Schrödinger’s FEP+ platform.

The open-source OPLS-AA is available in GROMACS and handles proteins adequately, though AMBER and CHARMM have larger protein validation datasets. Where OPLS-AA genuinely shines is for systems where organic solvent properties matter — solubility studies, partition coefficients, and binding in non-aqueous environments.