Driven by the constant increase in computational resources and power, modeling of biomolecular systems is now more accessible than ever. As simulation time and system size barriers are continuously pushed, the proper treatment of the underlying physics may become the bottleneck in improving our understanding of the systems of interest. Hence it seems to be an ideal time to utilize the increased computational power to improve force field parameters and look at their functional form as well. Starting from a non-polarizable force field, explicit polarizability was added to all heavy atom sites and subsequently the van der Waals parameters were recalibrated in Chapter 2. Only by defining a very systematic grid interpolation method to search local space, sets of parameters (34) could be found that properly describe the physio- chemical properties of the series methanol to butanol. With heats of varporization (∆Hvap) and densities (ρ) within 1.5% from experiment these models describe the thermodynamic poperties of the training set well. In Chapter 3 we explored the possibility of fitting dispersion parameters directly from QM calculations in an attempt to better calibration starting points. For a set of 11 linear and branched alkanes, dispersion parameters were computed using exchange-hole-dipole moment (XDM) calculations. As these calculations require atoms-in-molecules (AIM) partitioning of the electron density, we tested four methods suggested in literature that could be incorporated into the XDM formalism. Of these four methods an iterative Hirshfeld scheme was found to be the most suitable in com- bination with the XDM calculations. The thermodynamic properties of the linear and branched alkanes could be described well using a dispersion potential that explicitly includes the C6 instantaneous dipole-dipole and C8 instantaneous dipole-quadrupole terms. In Chapter 4 we investigated if use of our dispersion potential with explicit higher- order terms translates to improved description of polar groups as well. We choose water as a first test case because properly describing its unique properties in the condensed phase (such as its density maximum at 277 K) may well require a balanced treatment of dispersion and electrostatic interactions during simulation. We found our model to describe pure-liquid properties and temperature response well, indicating also the promise of higher-order dispersion potentials for polar compounds. We introduced a consensus-fitting approach as an extension to our previous (non-consensus) QM/MM fitting protocol, allowing us to find values for polarizabilities that are consistent between independent sets of hydration shells. After showing in Chapter 4 that this new way of fitting polarizabilities works well for our water model, we extended the fitting to other biomolecular building blocks in Chapter 5, in which results for a broad collection of both apolar and polar small molecules is presented. In this chapter we showed that indeed the original approach performs poorly (in terms of uncertainty) on semi-polar compounds, where the effective electric fields can be small and inconsistent. Following the consensus approach these uncertainties are significantly reduced. The work in Chapter 6 utilizes the methods presented in Chapters 2 to 5 to calibrate a force field for a series of 49 organic molecules, including 35 polar and 14 apolar compounds, which can serve as amino-acid side-chain analogs and other relvant biomolecular building blocks. Using the XDM determined values for C6 and C8 dispersion and our QM/MM determined polarizabilities, only six free parameters were left. These free parameters represent a free-state atomic radius and were rescaled for each van der Waals type according to the volume of the partitioned density, as com- pared to the free state. This resulted in a well-balanced force field for condensed liquid simulation that predicted densities and heats of vaporization well.
|Award date||11 Mar 2021|
|Place of Publication||s.I.|
|Publication status||Published - 11 Mar 2021|