The force field jungle

Even though in our lab most work is done with or on the coarse grain Martini force field, atomistic force fields are also important in our daily work. Beit as a basis/benchmark for the Martini force field, for the use in multiscale simulations, or for some interesting projects of their own. There are several 'major' atomistic (protein) force fields around which all have many different version, each with their own strengths an weaknesses. Recently several papers were published that systematically compared a range of different force fields, trying to find out which one currently is the "King of the Jungle". What follows is a mini-review of those papers.

Best et al: Are Current Molecular Dynamics Force Fields too Helical?, 2008 (
The authors compare the Ramachandran space sampled by a Ala5 peptides simulated using several different force fields (Amber03, Amber99SB, AmberGS, CHARMM27/cmap, OPLS-aa/L, Gromos53a6 and Gromos43a1) to J-coupling NMR experiments. Conclusions: in general the force field do sample the alpha-helical region to much, they sample Ramachandran space very differently, the best force fields agree reasonably with NMR experiments. The best scoring force fields are Gromos43A1, Amber03, OPLS-aa/L and CHARMM27/cmap.

Lange et al: Scrutinizing Molecular Mechanics Force Fields on the Submicrosecond Timescale with NMR Data, 2010 (
The authors compare microsecond long simulations of Ubiquitin and the gb3 domain of Protein G for 6 different force fields (Amber03, Amber99SB, OPLS/AA, Charmm22, Gromos43a1, Gromos53a6) to NMR data (RDC, 3h-J-couplings, NOE). Conclusions: PME treatment off the electrostatics gives better results compared to reaction-field/cutoff. Large differences between FFs. Simulations longer then ~100ns have an increased probalitiy of sampling nonnative conformational states. Hydrogen bonds are too weak. Best forcefield: AMBER99sb.

Beauchamp et al: Are Protein Force Fields Getting Better? A Systematic Benchmark on 524 Diverse NMR Measurements, 2012 (
Several new (updates to) force fields were available, motivating the authors to test 11 FFs ((Amber96, Amber99, Amber03, Amber03*, Amber03w, Amber99sb*, Amber99sb-ildn, Amber99sb-ildn-phi, Amber99sb-ildn-NMR, CHARMM27, OPLS-AA) using 5 different solvent models (GBSA, TIP3P, SPC/E, TIP4P-EW, TIP4P/05) using simulations of dipeptides, tripeptides, tetra-alanine, and ubiquitin). Conclusions: Although the differences between the explicit solvent models are small, TIP4P-EW generally gives the best performance. Don't use Amber 96, Amber99 or OPLS-AA. Best forcefields: Amber99sb-ildn-phi, ff99sb-ildn-NMR ('ildn' indicate the set of improved sidechain torsion angles and 'phi' or 'NMR' indicate two different sets of backbone parameters).

Cino et al: Comparison of Secondary Structure Formation Using 10 Different Force Fields in Microsecond Molecular Dynamics Simulations, 2012 (
Simulations of a hairpin forming peptides in 10 different force fields (Amber ff99SB-ILDN, Amber ff99SB*-ILDN, Amber ff99SB, Amber ff99SB*, Amber ff03, Amber ff03*, GROMOS96 43a1p, GROMOS96 53a6, CHARMM27, and OPLS-AA/L). Conclusions: OPLS-AA and (to a lesser extent) CHARMM-27 don't fold to a hairpin. The other force field perform all reasonably, however Gromos (both 53a6 and 43a1p) perform best on most measures, 43a1p slightly better even.

Lindorff-Larsen et al: Systematic Validation of Protein Force Fields against Experimental Data, 2012 (
Comparison of 8 FFs (Amber99SB-ILDN, Amber99SB*-ILDN, Amber03, Amber03*, OPLS-AA, CHARMM22, CHARMM27, CHARMM22*). Simulations of ubiquitin and the gb3 domain compared to NMR scalar couplings, order parameters and RDCs. Helical/sheet propensity of small peptides as a function of temperature compared to circular dichroism and abillity to fold the Villin headpiece and the WW domain. Conclusions: FFs are getting better over time. Don't use Charmm22, OPLS-AA, CHARMM-27. Use Amber99FB*-ildn or CHARMM22*.

Shirts et al: Extremely precise free energy calculations of amino acid side chain analogs: Comparison of common molecular mechanics force fields for proteins, 2003 (
This paper is clearly older than the rest, but is interesting due to its very different approach. The authors have calculated "the free energy of hydration of 15 amino acid side chain analogs derived from recent versions of the OPLS-AA, CHARMM, and AMBER parameter sets in TIP3P water using thermodynamic integration." Since the van der Waals and Coulomb paramters are largely unchanged in the force fields, these test will still be applicable to the newest versions of the FFs.Conclusions: The RMSD compared to experiment are rather high, all FFs and most AAs being not soluble enough. The differences between AAs in a FF are rather good. The RMSD over all AA for AMBER, CHARMM and OPLS-AA are: 1.35 kcal/mol, 1.31 kcal/mol, and 0.85 kcal/mol.

General Conclusions
Protein force fields are getting better! Different authors mention that the estimated error for the newest force fields is of the same order as the error in the experimental (NMR) values they use. There are different experimental values to test against and there are still plenty of opportunties to test the force fields (so, if you are bored...)

Which FF is the best depends on what you find important. The large amount of development that recently went in to Amber makes it generally good (Amber99SB-ildn possibly with the phi/NMR corrections). It is a pity that the recent new version of the Gromos FF (54A7 ( and 54A8 ( were not part of any of the tests (and Gromos was not part of more of the tests) since, especially the 43A1, FF performed rather good where it was included.

Different author suggest future developments, amongst which the improvement of hydrogen bonding capabilities, reparameterization of VDW and partial charges. Another exciting development I'm interested to see what it will improve are the polarizable force fields. And quite important are other molecules for which a forcefield has parameters: ligands, DNA and lipids. I know recently a few papers comparing lipid forcefields have been published. Although none of them seemed as extensive as the above papers, it will  be interesting to see what their conclusions are.