Bringing the MMFF force field to the RDKit: implementation and validation
© Tosco et al. 2014
Received: 12 March 2014
Accepted: 30 June 2014
Published: 12 July 2014
A general purpose force field such as MMFF94/MMFF94s, which can properly deal with a wide range of diverse structures, is very valuable in the context of a cheminformatics toolkit. Herein we present an open-source implementation of this force field within the RDKit. The new MMFF functionality can be accessed through a C++/C#/Python/Java application programming interface (API) developed along the lines of the one already available for UFF in the RDKit. Our implementation was fully validated against the official validation suite provided by the MMFF authors. All energies and gradients were correctly computed; moreover, atom type and force constants were correctly assigned for 3D molecules built from SMILES strings. To provide full flexibility, the available API provides direct access to include/exclude individual terms from the MMFF energy expression and to carry out constrained geometry optimizations. The availability of a MMFF-capable molecular mechanics engine coupled with the rest of the RDKit functionality and covered by the BSD license is appealing to researchers operating in both academia and industry.
KeywordsMolecular mechanics Force field MMFF RDKit
Molecular mechanics force fields are the workhorse of computational chemists for molecular simulations, owing to their low computational demands compared to CPU-intensive quantum mechanical methods. Drug designers and cheminformaticians are typically most interested in general purpose force fields, namely those which deliver good performance over a broad spectrum of different structures, ranging from biological macromolecules to small drug-like molecules. Such force fields are ideal to optimize 3D conformations of ligands or receptor/ligand complexes in batch workflows, because they seldom fail due to missing parameters. Examples of generalized force-fields include UFF , MMFF94 -, OPLS ,, GAFF  and CGenFF .
Among these, one of the most widely used and appreciated is MMFF94, which was developed about twenty years ago by Halgren at Merck Research Laboratories and documented by a series of five papers on J. Comput. Chem. -. A couple of years later, MMFF94 was complemented by the “static” MMFF94s variant , which is characterized by a different parameterization of torsional and out-of-plane interactions geared towards geometry optimization studies (MMFF94 was recommended for molecular dynamics simulations). For the remainder of this paper, “MMFF” will collectively refer to both MMFF94 and MMFF94s variants. MMFF has shown robustness and quality in dealing with small molecules superior to the majority of generalized force fields in use at the time of its first introduction ,, and also on par or better than the more recent OPLS2005 and GAFF force fields . Its versatility and effectiveness warranted its implementation in all major commercial molecular modelling packages (CCG MOE, Schrödinger Maestro, MolSoft ICM, Certara SYBYL-X) and cheminformatics toolkits (ChemAxon, OpenEye) currently on the market.
Herein we present an implementation of MMFF within the open-source cheminformatics toolkit RDKit ; MMFF functionality can be accessed through C++, C#, Python and Java application programming interfaces (APIs). While other non-commercial implementations of MMFF do exist -, ours is a complete implementation according to the definition given by Kearsley ,, and, to the best of our knowledge, the only such one available under the permissive BSD 3-clause license.
The first step in building the force field for a given molecular system is assigning correct types to each atom. MMFF identifies 216 symbolic atom types, based on the chemical nature and environment of each atom, which are encoded into 95 numeric atom types. Correct atom typing critically depends on the proper attribution of the aromaticity flags. Since the aromaticity model used by MMFF differs from the one normally used throughout the RDKit, aromaticity has to be re-perceived according to MMFF criteria starting from a kekulized representation of the molecule. Subsequently, atom types are assigned to heavy atoms followed by hydrogens.
As a second step, atom-centered partial charges are computed according to the MMFF charge model , which requires assignment of formal charges. The MMFF formal charge paradigm is based on resonant charges distributed over heteroatoms of the respective functional groups, which is different to the one implemented in the RDKit. In the API, atom types and charges are assigned upon construction of an instance of the MMFFMolProperties class. This class includes methods to choose between the MMFF94/MMFF94s variants, to access atom types and partial charges, to set the dielectric model (constant or distance-dependent) and the dielectric constant, or to exclude selected terms from the energy expression (Equation 1), respectively.
By calling the constructForceField() function, or its Python counterpart MMFFGetMoleculeForceField(), all bonded and non-bonded interactions in the molecular system under study, depending on its structure and connectivity, are loaded into the energy expression. Force constants and equilibrium values for bonded interactions are retrieved from the tables attached to the MMFF papers  by means of a binary search algorithm as recommended by Halgren. We also tried the map object implemented in the standard C++ library, which indexes database elements through hash tables, but it proved slightly slower. For interactions lacking a specific parameterization, a staged “step-down” procedure is carried out, in which increasingly generic values are sought .
In this case, the constraint is applied whenever the atom in question moves farther than rmax from its original position. The possibility of adding these restraining potentials to the force field expression has also been added to the existing RDKit UFF implementation.
Once all bonded and non-bonded interactions, plus optional restraints, have been loaded into the MMFF energy expression, potential energy and gradients of the system under study can be computed or minimized via RDKit’s ForceFields::ForceField::calcEnergy(), ForceFields:: ForceField::calcGrad() and ForceFields::ForceField::minimize() methods. A few examples of simple and constrained geometry optimizations and potential energy calculations are available as Additional files 1, 2 and 3; more examples can be found in the RDKit test suite (both C++ and Python).
Results and discussion
correct assignment of atom types and charges;
enumeration of bonded and non-bonded interactions and correct assignment/calculation of equilibrium values and force constants;
correct calculation of the energy contribution for each interaction.
Our implementation passes all validation tests for both MMFF94 and MMFF94s suites; hence, it can be labelled as a complete MMFF implementation according to the cited criteria given by Kearsley ,. The validation C++ code is included as part of the standard RDKit test suite, and is also available as Additional files 1, 2 and 3. On top of the official validation, we performed three additional robustness tests. Firstly, since the OPTIMOL log file does not include per-atom energy gradients, we checked them against those computed by the MMFF implementation available in TINKER , and they proved to be identical. Secondly, we converted the 3D structures provided in the MMFF validation suites into SMILES strings using the MolToSmiles() RDKit function, and then rebuilt 3D coordinates with DGeomHelpers::EmbedMolecule(); the raw 3D geometries were finally optimized with our MMFF implementation (see Additional files 1, 2 and 3). Obviously, since now the rebuilt conformations were different from those in the validation suite, energies could not be compared anymore. Still, all atom types, charges and force constants were correctly assigned. This proves the robustness of our assignment algorithms against variations of bond orders due to degenerate kekulization of condensed aromatic systems. Finally, upon suggestion of a reviewer, we challenged the atom typing code against 100 random shuffles of the atom order in the input coordinate files, obtaining correct results in all cases.
We presented a complete and validated implementation of the MMFF94/MMFF94s force fields within the open-source cheminformatics toolkit RDKit. The integration with the RDKit, which is licensed under a 3-clause BSD license, makes this implementation appealing to both academic and industrial users. The MMFF-related C++, C#, Python and Java APIs have the same architecture as those previously implemented for UFF; hence, it is straightforward to switch existing programs and scripts which relied upon the RDKit UFF force field implementation to MMFF. A comprehensive application to computer-aided drug design of the MMFF implementation described herein will be the subject of a forthcoming paper.
Availability and requirements
The MMFF implementation object of this work is included in the BSD-licensed open source toolkit RDKit  since Release_2013.09.1, and can be accessed from C++, C#, Python and Java applications.
PT thanks the Novartis Institutes for BioMedical Research for supporting his 3-month stay as a visiting scientist under the terms of a consultancy agreement.
- Rappe AK, Casewit CJ, Colwell KS, Goddard WA, Skiff WM: UFF, a full periodic table force field for molecular mechanics and molecular dynamics simulations. J Am Chem Soc. 1992, 114: 10024-10035. 10.1021/ja00051a040.View ArticleGoogle Scholar
- Halgren TA: Merck molecular force field. I. Basis, form, scope, parameterization, and performance of MMFF94. J Comput Chem. 1996, 17: 490-519. 10.1002/(SICI)1096-987X(199604)17:5/6<490::AID-JCC1>3.0.CO;2-P.View ArticleGoogle Scholar
- Halgren TA: Merck molecular force field. II. MMFF94 van der Waals and electrostatic parameters for intermolecular interactions. J Comput Chem. 1996, 17: 520-552. 10.1002/(SICI)1096-987X(199604)17:5/6<520::AID-JCC2>3.0.CO;2-W.View ArticleGoogle Scholar
- Halgren TA: Merck molecular force field. III. Molecular geometries and vibrational frequencies for MMFF94. J Comput Chem. 1996, 17: 553-586. 10.1002/(SICI)1096-987X(199604)17:5/6<553::AID-JCC3>3.0.CO;2-T.View ArticleGoogle Scholar
- Halgren TA, Nachbar RB: Merck molecular force field. IV. Conformational energies and geometries for MMFF94. J Comput Chem. 1996, 17: 587-615.Google Scholar
- Halgren TA: Merck molecular force field. V. Extension of MMFF94 using experimental data, additional computational data, and empirical rules. J Comput Chem. 1996, 17: 616-641. 10.1002/(SICI)1096-987X(199604)17:5/6<616::AID-JCC5>3.0.CO;2-X.View ArticleGoogle Scholar
- Halgren TA: MMFF VI. MMFF94s option for energy minimization studies. J Comput Chem. 1999, 20: 720-729. 10.1002/(SICI)1096-987X(199905)20:7<720::AID-JCC7>3.0.CO;2-X.View ArticleGoogle Scholar
- Halgren TA: MMFF VII. Characterization of MMFF94, MMFF94s, and other widely available force fields for conformational energies and for intermolecular-interaction energies and geometries. J Comput Chem. 1999, 20: 730-748. 10.1002/(SICI)1096-987X(199905)20:7<730::AID-JCC8>3.0.CO;2-T.View ArticleGoogle Scholar
- Jorgensen WL, Maxwell DS, Tirado-Rives J: Development and testing of the OPLS all-atom force field on conformational energetics and properties of organic liquids. J Am Chem Soc. 1996, 118: 11225-11236. 10.1021/ja9621760.View ArticleGoogle Scholar
- Kaminski GA, Friesner RA, Tirado-Rives J, Jorgensen WL: Evaluation and reparametrization of the OPLS-AA force field for proteins via comparison with accurate quantum chemical calculations on peptides. J Phys Chem B. 2001, 105: 6474-6487. 10.1021/jp003919d.View ArticleGoogle Scholar
- Wang JM, Wolf RM, Caldwell JW, Kollman PA, Case DA: Development and testing of a general amber force field. J Comput Chem. 2004, 25: 1157-1174. 10.1002/jcc.20035.View ArticleGoogle Scholar
- Vanommeslaeghe K, Hatcher E, Acharya C, Kundu S, Zhong S, Shim J, Darian E, Guvench O, Lopes P, Vorobyov I, Mackerell AD: CHARMM general force field: a force field for drug-like molecules compatible with the CHARMM all-atom additive biological force fields. J Comput Chem. 2010, 31: 671-690.Google Scholar
- Gundertofte K, Liljefors T, Norrby PO, Pettersson I: A comparison of conformational energies calculated by several molecular mechanics methods. J Comput Chem. 1996, 17: 429-449. 10.1002/(SICI)1096-987X(199603)17:4<429::AID-JCC5>3.0.CO;2-W.View ArticleGoogle Scholar
- Hornak V: Comparison of robustness and quality of three small molecule force-fields. Proceedings of CUP XI, Santa Fe, NM, USA. 2010Google Scholar
- RDKit: Cheminformatics and Machine Learning Software. ᅟ. 2013, ᅟ: ᅟ-[http://www.rdkit.org], [http://www.rdkit.org]
- O’Boyle NM, Banck M, James CA, Morley C, Vandermeersch T, Hutchison GR: Open Babel: an open chemical toolbox. J Cheminform. 2011, 3: 33-10.1186/1758-2946-3-33.View ArticleGoogle Scholar
- TINKER-Software Tools for Molecular Design. [http://dasher.wustl.edu/ffe/], [http://dasher.wustl.edu/ffe/]
- MCCCS Towhee. [http://towhee.sourceforge.net/], [http://towhee.sourceforge.net/]
- Chemkit. [http://wiki.chemkit.org/Main_Page], [http://wiki.chemkit.org/Main_Page]
- Kearsley S: MMFF94 Validation Suite. [http://ccl.net/cca/data/MMFF94/index.shtml], [http://ccl.net/cca/data/MMFF94/index.shtml]
- Kearsley S: MMFF94s Validation Suite. [http://ccl.net/cca/data/MMFF94s/index.shtml], [http://ccl.net/cca/data/MMFF94s/index.shtml]
- Halgren TA: Merck molecular force field. I. Basis, form, scope, parameterization, and performance of MMFF94. J Comput Chem. 1996, 17: 490-519. 10.1002/(SICI)1096-987X(199604)17:5/6<490::AID-JCC1>3.0.CO;2-P. Supplementary material [ftp://ftp.wiley.com/public/journals/jcc/suppmat/17/490/MMFF-I_AppendixB.ascii], [ftp://ftp.wiley.com/public/journals/jcc/suppmat/17/490/MMFF-I_AppendixB.ascii]View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.