Molecular Mechanics (MM) simulations are useful tools for a broad range of applications, including protein folding and structure, materials research, and determination of liquid structure and properties. Such simulations are also popular choices to measure binding thermodynamics of drug-protein interactions, with implications in drug development. Advances in hardware and algorithms have greatly improved our ability to converge computed properties to good precision, however, achieving high accuracy relative to experimental observables will require further improvements in the potential functions, or force fields, used to calculate the energies and atomic forces. Historically, some of the most successful force fields derive key parameters by training against experimental data, such as heats of vaporization and densities of neat liquids of small organic compounds. While there has been some success in parameterization along these lines, it remains an often difficult task to sufficiently parameterize large portions of chemical space necessary for most biomolecular simulations. Another possible avenue for parameter generation is quantum mechanical data on the molecules to be studied; in theory, the information obtained from a given electron density can be used to assign parameters on a more physically sound basis.<br/>We have developed a methodology that takes information contained in electron densities of molecules and assigns unique force-field parameters to each atom in the molecule. One consideration of this approach is the necessity of “mapping parameters”, since the ideal mapping of QM information to force-field parameters is not known. We choose to keep the number of parameters small, only having two mapping parameters for each element (with the exception of hydrogen, where polar hydrogens have their own unique mapping parameters). We tune these parameters so that they best replicate an informative set of experimental observables. Preliminary results from the application of this approach to the assignment of Lennard-Jones non-bonded parameters has been met with comparable results as a current popular force-field GAFF, despite fewer adjustable parameters and a smaller set of training data. This method has potential to generate force-fields which are informed by quantum mechanical data while remaining with only a few adjustable parameters, making the optimization much more tractable and, subsequently, an improvement in simulation accuracy.

Towards improved accuracy in calculation of binding thermodynamics

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.