Skip to content

Small molecule forcefield

John Chodera edited this page Jun 25, 2014 · 7 revisions

Working group

  • (JDC) John D. Chodera, MSKCC <john.chodera@choderalab.org>
  • (KAB) Kyle A. Beauchamp, MSKCC <kyle.beauchamp@choderalab.org>
  • (ABR) A. Bas Rustenburg, MSKCC <bas.rustenburg@choderalab.org>
  • (PBG) Patrick B. Grinaway, MSKCC <patrick.grinaway@choderalab.org>
  • (LPW) Lee-Ping Wang, Stanford <leeping@stanford.edu>
  • (CIB) Christopher I. Bayly, OpenEye <bayly@eyesopen.com>
  • (DLM) David L. Mobley, UCI <dmobley@uci.edu>
  • (DM) Daniel McKay, Novartis <daniel.mckay@novartis.com>
  • (PL) Paul Labute, CCG <paul@chemcomp.com>

Desiderata

  • start with neat and mixed solvents
  • simultaneous fitting of all solvents
  • bond charge increment model
  • sigma hole on C-Cl / bond-centered charges in general, judiciously chosen
  • MRS has reweighting scheme for bond length changes
  • scaling factors for torsions an adjustable parameter
  • bond lengths and valence angles aren't going to make a huge difference (but maybe want them close to some QM ref for good QM/MM overlap?)
  • allow decoupling of parameter types for nonbonded, bonds, angles, torsions: implement with combinatorial atom types

Action items

QM torsion profiles

  • Discussion with CIB will produce an initial list of small molecules with one or two rotors for which QM torsion profiles are to be computed by JDC.
  • We will use LPW's crank tool to produce QM torsion profiles.
  • Torsions are driven on a 10-degree grid while other degrees of freedom are optimized.
  • We will use LPW's preferred QM level of theory (TBD) and protocol

Measurement of densities of neat liquids and binary mixtures

The Chodera lab will measure densities of neat liquids and binary mixtures using the Mettler-Toledo DM40 density meter capable of measurements in the 0-3 g/cm3 range with 0.0001 g/cm3 accuracy over the temperature range 0-91 C. We will use a Mettler-Toledo SC30 sample changer to automate sample collection.

  • JDC will send CIB a list of liquids we can inexpensively purchase and handle safely.
  • CIB will select a subset of liquids that will provide a useful "universe" of initial measurements.
  • Each neat liquid will have densities measured for at least 30 temperatures from 0-91 C, avoiding freezing/boiling ranges.
  • Each binary mixture will have 30 mole fractions x 30 temperatures = 900 measurements, collected in roughly 8 hours. We will avoid freezing/boiling temperature ranges.
  • The Mettler-Toledo Quantos liquid dispensing system will be used to prepare samples, so the mole fractions measured will not always be the same, but will be accurately known.
  • Both dispensing and data collection will be automated by Mettler-Toledo LabX software.
  • Raw and processed data will be checked into the GitHub. Processed data should be amenable to input into ForceBalance.

Initial parameter fitting experiments

  • We will use some combination of QM and experimental data to reparameterize a GAFF-like forcefield on a subset of chemical space.
  • We can use this opportunity to experiment with Bayesian atom type selection schemes
  • The exact form of the forcefield remains to be worked out, but multiple experiments could be attempted within ForceBalance

How parameter assignment might work

Charges

  • A scheme is used to rapidly enumerate a number of small-molecule conformations (e.g. omega)
  • These conformations are refined to ensure there are no intramolecular contacts (e.g. by minimizing with absolute values of MMFF charges)
  • For each conformation, a semiempirical QM method (e.g. AM1 or PM3) is used to compute an initial CM2 population analysis
  • A series of bond charge corrections are applied to refine the small molecule charges
  • The charges are averaged over conformations
  • Charges for symmetry-equivalent atoms (e.g. methyl protons) are averaged

Atom types

  • A SMARTS-based atom typing system is used to define atom types
  • A hierarchical scheme is used for atom typing, with more general types occurring first and more specific types overriding them
  • Lennard-Jones (or exp-6, or whatever) parameters are assigned based on atom types

Bond types

  • A SMARTS-based bond typing system is used; bond types may not correspond to atom types
  • A hierarchical scheme is used for typing
  • Harmonic bond parameters are assigned
  • Some bonds may also have virtual sites, such as C-Cl "sigma hole" point charges. Bond charge corrections are used to partition charges to these virtual sites from their parent bond atoms.

Angle types

  • A SMARTS-based bond typing system is used; angle types may not correspond to atom types
  • A hierarchical scheme is used for typing
  • Harmonic angle parameters are assigned

Torsion types

  • A SMARTS-based bond typing system is used; torsion types may not correspond to atom types
  • A hierarchical scheme is used for typing
  • A variable number of torsion parameters are assigned; CMAP-style torsions may not be necessary

What parameters might be optimized

  • bond charge corrections (the number of corrections, SMARTS bond types, and charge correction)
  • bond charge corrections for creating bond midpoint charges (the number of corrections, SMARTS bond types, virtual site position along bond, and charge correction to move charge from bonded atoms to virtual site)
  • vdW interaction type (Lennard-Jones, Exp-6, other), SMARTS atom types, parameters, and mixing rules
  • bond types (SMARTS bond types, force constant, equilibrium length)
  • angle types (SMARTS angle types, force constant, equilibrium angle)
  • torsion types (SMARTS torsion types, Fourier expansion terms)