This provides the first general-purpose implementation of a SMIRKS Native Open Force Field (SMIRNOFF) created by the
Open Force Field Initiative.
You can parameterize small molecules with SMIRNOFF using the
ForceField class in the openforcefield toolkit
for simulations with OpenMM.
Details about this new format are documented in our recent publication (doi:10.1021/acs.jctc.8b00640 or bioRxiv).
Usage examples can be found in the openforcefield repository.
DOIs for each force field in this repository can be found in the following table:
| Filename | DOI |
|---|---|
| smirnoff99Frosst-1.1.0.offxml |  |
| smirnoff99Frosst-1.0.9.offxml |  |
| smirnoff99Frosst-1.0.8.offxml |  |
| smirnoff99Frosst-1.0.7.offxml |  |
| smirnoff99Frosst-1.0.6.offxml |  |
| smirnoff99Frosst-1.0.5.offxml |  |
| smirnoff99Frosst-1.0.4.offxml |  |
| smirnoff99Frosst-1.0.3.offxml |  |
| smirnoff99Frosst-1.0.2.offxml |  |
| smirnoff99Frosst-1.0.1.offxml |  |
| smirnoff99Frosst-1.0.0.offxml | Same as 1.0.1 |
conda install -c omnia smirnoff99frosst(SMIRNOFF99frosst was formerly known as SMIRFF99frosst)
Installing this package exposes an entry point that makes the smirnoff99frosst/offxml directory easily accessible by other packages in the same python installation. If the Open Force Field toolkit is installed, it will automatically detect and use this entry point:
>>> from openforcefield.typing.engines.smirnoff import ForceField
>>> ff = ForceField('smirnoff99Frosst-1.0.9.offxml')
Otherwise, the entry point can be accessed by querying the openforcefield.smirnoff_forcefield_directory entry point group.
>>> from pkg_resources import iter_entry_points
>>> for entry_point in iter_entry_points(group='openforcefield.smirnoff_forcefield_directory'):
... print(entry_point.load()())
The provided OFFXML (force field) files are successive versions of a general-purpose small molecule force field, written in the SMIRNOFF format; this force field should cover all or almost all of drug-like chemical space, and illustrate some of the major functionality of the SMIRNOFF format as well as how it simplifies the specification of force field parameters in a compact and chemically sensible way.
HOWEVER, this is not expected to be (at present) an especially accurate small molecule force field. Its authors (see History, below) expect that while coverage will initially be good, additional refinements will be required (and possibly some expansion of the number of parameters) before it can rival current force fields such as GAFF or OPLS in accuracy. However, we are optimistic that it already rivals them in extensibility, and potentially with relatively minimal work can be extended to be a compelling present-day small molecule force field.
As with typical members of the AMBER force field family, SMIRNOFF99Frosst is intended to be used with RESP or AM1-BCC charges which are not specified by the force field itself.
Differences from parm99 and parm@frosst:
SMIRNOFF99Frosst is neither parm99 nor parm@frosst exactly, for a number of reasons including that:
- It covers, or should cover with only slight modification, all reasonable organic chemistry, which neither of the above did
- It is much simpler (because of parameter consolidation by SMIRKS, because of removing questionably differentiated parameters, and because its goal is to be a "good starting point" than a finished product)
- It introduces a number of approximations/consolidations for conciseness/clarity
- It fixes a number of mistakes/omissions, etc.
The second and third bullet points are particularly important. When many parameters (such as torsional barrier heights, or equlibrium angles) for similar chemical functionality differed only slightly in the existing force fields, these were often deliberately consolidated to have a single value.
The last bullet point is also noteworthy. Because of limitations of atom typing, many torsions were missing in parm99/parm@frosst, resulting in generics being erroneously applied when a more specialized torsion could have served. The use of SMIRKS typically means that this is not the case, so many mistakes are eliminated.
This forcefield, SMIRNOFF99Frosst, is a logical descendant of AMBER's parm99 forcefield as well as Merck-Frosst's parm@frosst, but is generalized/simplified and put into the SMIRNOFF format.
SMIRNOFF99Frosst was created by Christopher I. Bayly (with help from Caitlin C. Bannan and David L. Mobley, UC Irvine) during his sabbatical at UCI during Summer 2016. It was created by hand curation of the original parm99 and parm@frosst parameter and frcmod files (manually creating SMIRKS patterns, condensing parameters, etc.). Usage of SMIRKS, along with application of Bayly's chemical intuition and extensive history with the AMBER force field family, allowed the number of parameter lines to be reduced from the multiple thousands of parm99+parm@frosst to the few hundred seen here.
We expect to continue to manually expand this to cover more chemical space and, when needed, to introduce additional parameters to improve accuracy. Subsequent versions are expected to build on Bayly's hand created version with some enhancements by Caitlin Bannan by improving specific parameters, further differentiating some chemical functionality, etc. However, derivatives are expected to retain this basic structure and the AMBER legacy/history; a full refitting of bonded parameters would result in a separate force field rather than a new version of this force field.
In general, this repository will only contain force fields that aim to reproduce the behavior of parm99/parm@Frosst using the SMIRNOFF format. Once the Open Force Field Initiative begins parameter optimizations, the resulting force field should be migrated to separate base name.
Applicable in general to SMIRNOFF-format FFs produced by the Open Force Field Consortium
Force fields moving forward will be called name-X.Y.Z
Xdenotes some major change in functional form.Yis the parameterization epoch / generation, or a minor change that can affect energy.Zis a bugfix version -- e.g. something we've caught and corrected.
In OpenMM, application of SMIRNOFF99Frosst to small molecules should be straightforward via openforcefield with examples available in the openforcefield repository.
Additionally, with ParmEd, it should be possible to convert parameterized OpenMM systems into other formats such as AMBER, CHARMM, or GROMACS, making this forcefield available in a variety of packages.
- Version 1.0/Version 1.0.1 (equivalent): Initial release after hand curation by C. I. Bayly and C. C. Bannan. DOI 10.5281/zenodo.154235
- Version 1.0.2: Fixes an out-of-order generic (bond
[#6X2:1]-[#6:2]) as per Issue 4. - Version 1.0.3: Bug fixes -- adding one omitted bond length, fixing four torsional smirks patterns, and adding one missing torsional term as detailed in smarty issue 164
- Version 1.0.4: Bug fixes -- #11: Fix the parm@Frosst-derived C-OS bond length so it does not also match (and thus override) C-OH, switching from SMIRKS of
[#6X3:1](=[#8X1])-[#8X2:2]to[#6X3:1](=[#8X1])-[#8X2H0:2], which avoids overriding[#6X3:1]-[#8X2H1:2]. And then add back in a generic which should have been present (PR 15, fixing a bug introduced by PR 11) - Version 1.0.5: Substantially improved coverage of chemical space via more general generics as well as a variety of new parameters introduced via generalization/estimation from other force fields such as GAFF/GAFF2. This release, this version covers an internal set of molecules from DrugBank filtered to remove metal atoms and to contain only compounds with less than 200 heavy atoms. Full documentation of changes is available here.
- Version 1.0.6: Added monovalent ion parameters (Joung/Cheatham) for TIP3P as default. Added angle parameters for cyclobutyl groups. Replaced
Rdecorators withxto guarantee compatibility between OpenEye toolkits and RDKit SMIRKS parsing. - Version 1.0.7: Add hydroxyl hydrogen radii (as per SMIRNOFF initial paper); remove generics with pure wildcards (not even elemental types).
- Version 1.0.8: Fix human error in hydroxyl hydrogens; fix bond parameters between hydrogen and divalent carbons (issue #81); and fixed SMIRKS patterns for angle parameters around trivalent carbon in 5-membered rings (issue #84).
- Version 1.0.9: Addresses issue 89: Fixes torsion t56, where SMIRKS
[!1:1]-[#7X4,#7X3:2]-[#6X4;r3:3]-[*:4]should be[!#1:1]-[#7X4,#7X3:2]-[#6X4;r3:3]-[*:4]. The first means "not an isotope with mass 1" (!1), but we intend for it to apply to all hydrogens, so it has been changed to!#1. - Version 1.1.0: Adds hydrogen bond constraints. Updates contents to SMIRNOFF 0.3 spec (spec change performed by OFF toolkit 0.4.0 reading smirnoff99Frosst-1.0.9, and then writing the
ForceFieldobject out in the 0.3 spec). Fixes several issues with FF hierarchy, SMIRKS patterns. This fixes some issues with coverage of certain areas of chemical space, and parameters which could never be utilized due to hierarchy issues. Full details available in openforcefield/openforcefield Issue 367.
Contributors to the relevant .offxml files include:
- Christopher I. Bayly (OpenEye Software/UC Irvine)
- Caitlin C. Bannan (UC Irvine)
- David L. Mobley (UC Irvine)
Special thanks go to John D. Chodera (MSKCC) for his initial implementation of openforcefield toolkits and the SMIRNOFF format.
Andrea Rizzi (MSKCC) and Jeff Wagner (OFF/UC Irvine) contributed to the Python and Conda infrastructure of this package.
Project based on the Computational Molecular Science Python Cookiecutter version 1.0.