MMFF94 Compliance Statement

1. It fully produces the "total and component energies" of the 761 molecule "MMFF94 Validation" suite as posted at http://ccl.osc.edu/cca/data/MMFF94/index.shtml" (June 1999) to an accuracy of .00005 kcal/mol (+-5.0e-5).
2. The full functional form of MMFF94 has been implemented. as found in J. Comp. Chem., (17) 1996, 490-641.
3. The step-down protocol for default parameter assignment of MMFF94 is fully utilized for atoms H, C, N, O, F, Si, P, S, Cl, Br and I. (All of the atoms parameterized in MMFF94.)
4. All non-bonded ions described in the MMFF94 papers are fully implemented.
   FE+2 FE+3 F- Cl- Br- Li+ Na+ K+ ZN+2 CA+2 CU+ CU+2 MG+2
5. It produces the same results for both 'standard' ways of drawing tetra-coordinated sulfur and phosphorus compounds. These are the 'dative' bonding (MMFF94_dative.mol2 file of the Suite) and the 'hypervalent' bonding (MMFF94_hypervalent.mol2). Our implementation also correctly gets most 'delocalized' bonding structures. For example; using the 'aromatic' bond to represent delocalized electrons for both aromatic systems and delocalized charged systems such as sulfates and guanidinium complexes. Some exceptions to this are discussed in point 'E' in the second section of this page.
6. It is possible to draw molecules in Spartan which are not covered in the MMFF94 parameterizations, e.g. Boron cages, organo-metallic compounds, and carbo-cations. When submitting these to the minimizer we use a series of rules based on MMFF94, UFF and our own templates. These are not valid tests of MMFF94, and the output mentions the unknown elements and new force-fields. (The force field is labeled "MMFF94 with extensions" as opposed to "MMFF94".) A discussion of these extensions is described elsewhere. One can use the 'PURE' keyword which will not allow extensions to the MMFF94 and fail on any molecule which contains atoms not mentioned in (C) and (D) above.

1. No VdW cutoff is used ('LONGVDW' keyword). By default the VdW forces begin to be phased out at 7.0 A and are zero at 9.5 A.
   This can affect the energy by +- 0.01 kcal/mol
   The worst case in the validation suite is the molecule labeled 'JANDOR' which has 53 atoms and 1111 non 1-4 VdW pairs.
2. For the dative (and delocalized) data sets a geometry optimization was performed. These data sets were generated from the '.mol2' file (as opposed to the '.mmd' file), and the precision of the coordinates in the mol2 format is not high enough for an accuracy of .0001 kcal for a number of strained systems in the suite.
3. Spartan arbitrarily chooses Cu and Fe ions to be +2. To calculate CU1PW1 and FE3PW3 correctly, the keywords 'FFHINT=CU1~~+1' and 'FFHINT=FE1~~+3' are needed, respectively.

4. There is an inconsistency in the Optimol implementation when it chooses the torsional bond type of certain C-N bonds. To emulate this behavior we need to add the keyword 'MMFF_CN_TORSION_FIX' to 5 molecules: CYGUAN01, DIVJUN, FOYMAH, FULRAF and VEWZOM.

   For example in FOYMAH:
   
   

   Clearly both Nitrogens are the same, due to delocalization of electrons. Yet, torsions of the form C-C-N-H are different in Optimol's implementation. We believe they should be the same. They are identical for bond-stretch and bend rules. In Optimol two (of the four) have a torsion type of '2', the others are '0'.

   Comparing the output describing these torsions for the two programs (Optimol and Spartan) we see:
   
   --- Optimol (edited from the Validation suite) ---
C5 C4 N1 H11 37-57-55-36 2 .179.931 0.000 0.000 4.800 0.000
C5 C4 N1 H21 37-57-55-36 2 ..-2.064 0.006 0.000 4.800 0.000
C5 C4 N2 H12 37-57-55-36 0 -179.840 0.000 0.000 10.000 0.000
C5 C4 N2 H22 37-57-55-36 0 ..-1.442 0.006 0.000 10.000 0.000
--- Spartan default (edited) ---
2: H21 - N1 - C4 - C5 ==> [ 0.000 4.800 0.000 ] [2] 3
4: H11 - N1 - C4 - C5 ==> [ 0.000 4.800 0.000 ] [2] 3
6: H22 - N2 - C4 - C5 ==> [ 0.000 4.800 0.000 ] [2] 3
8: H12 - N2 - C4 - C5 ==> [ 0.000 4.800 0.000 ] [2] 3
--- Spartan with the MMFF_CN_TORSION_FIX keyword (edited)
2: H21 - N1 - C4 - C5 ==> [ 0.000 4.800 0.000 ] [2] 3
4: H11 - N1 - C4 - C5 ==> [ 0.000 4.800 0.000 ] [2] 3
6: H22 - N2 - C4 - C5 ==> [ 0.000 10.000 0.000 ] 3
8: H12 - N2 - C4 - C5 ==> [ 0.000 10.000 0.000 ] 3

   Another example is 'CYGUAN01'
   
   

   For the torsion C6-N1=C3-N3 and C1-N1=C3-N3 a torsion of type '2' is found. After 2 "rule-fall-backs" Spartan returns a force-constant of V2=4.8, which is the same that both programs get for C6-N1-C3-NH2. However, Optimol chooses a torsion type of '0' and V2=10.0.

   Note that this difference is usually undetectable as the NH2 group is nearly planar and differences only show up in strained systems.

   If the 'MMFF_CN_TORSION_FIX' keyword is not applied, the energy differences between the 'mmd' structures and the unminimized Spartan energies for the 5 molecules in question are:

   Molecule	Published Energy	Energy Without "Torsion Fix"	Delta	Energy With "Torsion Fix"
   VEWZOM	-9.37968	-9.379853	.00017	-9.379675
   FULRAF	98.97744	98.976814	.00062	98.977440
   FOYMAH	-4.87726	-4.880598	.00333	-4.877264
   DIVJUN	86.58284	86.577870	.00497	86.582836
   CYGUAN01	-254.74397	-254.773534	.02956	-254.74396

5. Spartan produces identical results for the dative and hypervalent test suites.

   We, (Wavefunction), have a number of customers who also use 'delocalized' bonds to represent delocalized charge and aromaticity. (i.e. Sybyl's type '5' bond, drawn with 2 lines, one solid and the other dashed.)

   Our goal is to get these representations to match correctly. This is problematic, especially in the case of fused rings as it confuses the 'upgrading-to-aromaticity' rules. For example; making sure the central ring in DAKCEX remains non-aromatic is difficult:
   
   

   Thus, we advise customers use the Keukule drawings of aromatic systems, unless one wants to specifically override the MMFF94 aromaticity rules.
   Other, non-aromatic delocalized charge distribution, (ie. SO2, NO2 and NCN+) should be ok.

Last modified: Thu Aug 16 20:49:25 GMT 2018