The following text has been reproduced from Cornell, et al., A Second Generation Force Field for the Simulation of Proteins, Nucleic Acids, and Organic Molecules, J. Am. Chem. Soc., vol 117, pp 5179-5197, 1995. Specifically, this is a summary of the text that appears in the Discussion section on pages 5192-5193. Edited text appears in italics.
"We will attempt to summarize the salient features of some of the more commonly used force fields here, in order to compare and contrast our approach (AMBER95) with theirs. They can be roughly grouped into four different categories, depending upon the nature and complexity of the force field equation:
The ECEPP force field of Scheraga(89) employs rigid internal geometries which allow a more efficient exploration of conformational space. This approach has the disadvantage that it can can cause certain conformations and conformational barriers to be too high in energy. A second force field which uses only partially rigid geometries is JUMNA(90), developed by Lavery and co-workers. This force field has been developed for nucleic acids and allows flexibility in the sugar ring but uses mainly internal geometries and keeps the bases rigid.
The SYBYL forcefield(91) has been developed for the calculation of internal geometries and conformational energies. Because it contains no electrostatic term, it is inappropriate for studying detailed condensed-phase properties. The YETI force field,(92) developed by Vedani and Huhta, is a modification of the Weiner, et al. force field with highly damped electrostatics and an angular dependent hydrogen bond (and metal ligation) potential added. This approach could be valuable in some modeling situations, where large and difficult to handle electrostatic energies are present, but it is also unlikely to be general and extendable to condensed-phase phenomena.
The category of simple diagonal force fields includes the Weiner, et al.,(5,6) GROMOS,(93) CHARMm,(94) and OPLS/AMBER(15) force fields. All of these force fields employ a simple harmonic diagonal representation for the bond and angle terms... The Weiner, et al., force field derived charges from fits to the electrostatic potential of a molecule whereas the other two force fields used empirical fits to interaction energies (CHARMm) or liquid and solid state data (GROMOS). The Weiner, et al., CHARMm, and GROMOS force fields all employ VDW parameters derived from crystal data, whereas the VDW parameters in the OPLS/AMBER and the CORNELL, et al.force fields are derived from liquid simulations... For heteronuclear interactions, the OPLS/AMBER and GROMOS force fields determine VDW interaction values using geometric mean combining rules. By comparison, Weiner, et al., Cornell, et al., and CHARMm employ arithmetic mean combining rules...GROMOS makes a further distinction of using different values for VDW parameters for a particular atom type, depending upon the second atom involved in the interaction. This has been shown to result sometimes in anomalous behavior.(95,96)
CHARMm94(99), AMBER, OPLS, and the Sun, et al.,(21) CFF91 forcefields reasonably reproduce condensed-phase properties. We wish to stress that both the OPLS and Sun, et al., parameters are appropriate and effective models to use in condensed-phase studies of organic molecules that are not highly strained or have very short nonbonded distances involving hydrogen.
While all five forcefields employ a simple Fourier expansion to represent the dihedral energy, some variation is also seen in the assignment of that energy...
Finally, the category of "more complex" force fields includes not only the MM2 and MM3 force fields for small molecules(2,3) but also two other force fields. These force fields go beyond the simple diagonal potential function in their inclusion of higher order terms as well as cross-terms for representing bonds and angles. The MM3 force field is the state-of-the-art for modeling organic molecules in the gas phase and has been carefully calibrated to reproduce many properties of these molecules. The focus of MM3 is quite different from that of the force field presented here (AMBER95) in that it is not oriented toward the representation of polar and ionic molecules in condensed phases, although, for example, some crystal minimizations were used to calibrate some of the nonbonded parameters. Its complex functional form is necessary for reproducing vibrational frequencies and subtlties of molecular geometries. The use of a 6-exponential nonbonded potential is more accurate than the 6-12 used here, particularly for close contacts such as those found in highly strained organic molecules. The MM2/MM3 model uses a point dipole approach for electrostatic interactions which has often worked well for modeling intramolecular properties but has not been rigorously established as a general model for modeling intermolecular interactions. MM2/MM3 has a large number of dihedral parameters specific to four-atom bond quartets which have been fit to a large set of data.
A second complex force field is the "Class II" one under development by Hagler and co-workers ESFF.(101) This force field has a functional form of similar complexity to that of MM2/MM3, but it differs in the extensive use of quantum mechanical energies and gradients for its calibration. The developers of this force field are pioneering new ways of deriving parameters and analyzing molecular interactions. The force field currently suffers, however, from the lack of a general charge model of the same caliber as the other parameters.
The third complex force field is the Merck Molecular Force Field (MMFF) under development by Halgren(102). The stated purpose of this force field is to be able to handle all fo the functional groups of interest in pharmaceutical design. The nonbonded function is a "buffered" 7-14 potential, which Halgren found to give the best fit to rare gas interactions, and an empirical bond dipole model is used to assign partial charges. The key calibration test set is a series of conformational energies calculated at a very high level of ab initio theory (MP4SDQ/TZP//MP2/6-31G*). Thus far, no condensed-phase simulations have been carried out, but they are planned. This approach has the advantage of generality to a large number of molecules, but at the expense of the use of a simple, empirical, generic charge model and a large number of dihedral parameters.