New theoretical techniques are being developed and characterized. These efforts are usually coupled with software development, and involve the systematic testing and evaluation of new ideas. Methods for evaluating long range interactions Long range electrostatic and dispersion interactions have important consequences for accurate modeling of macromolecular structures, particularly in highly anisotropic systems. After preliminary results showed that long range effects are crucial for accurately modeling lipid models, a large scale collaboration isunderway to update the CHARMM force fields, particularly those forlipids, supplanting distance based truncation methods with the particle mesh Ewald(PME)method. Becausevery few programshave PME capabilities for dispersion, we have developed a standalone library to enable these calculationsto be carried out by any program and have made it freely available to all researchers. The code uses state-of-the-art techniques to parallelize the calculationacross supercomputers and ongoing work is extending themultithreading capabilities to use graphics cards toaccelerate the calculations, as well as efficiently leveraging the large vector capabilities of modern processors. We have also developed a fresh approach for PME that can be more efficiently parallelized across nodes within a supercomputer. The bottleneck in such computations is due to the need to communicate large quantities of information between nodes. Our novel approach introduces a strategy to a priori compress the information, which should greatly reduce the communication bottleneck. This new algorithm will also be able to utilize the latest features of modern graphics cards and chips that were introduced to aid the machine learning community; work is underway to write PME code exploiting these features. We are also making these developments available to the community through our open source PME library. Polarization approches for multiscale modeling The AMOEBA polarizable force field has an elaborate functional form that permits very high accuracy descriptions of the complex environments introduced at interfaces and surrounding highly charged moieties. Our preliminary work has revealed that the polarization groups, introduced by AMOEBAs creators to ensure transferability, introduce problems related to efficiency and they prevent the simple formation of a variational quantum mechanical framework, which is a long standing goal of ours. We have tested a number of schemes for eliminating these polarization groups and are working with the creators of AMOEBA to incorporate these schemes into the emerging next-generation polarizable models. Obtaining QM/MM free energies via reparameterization of classical force fields with force matching We demonstrate that obtaining free energies at a high (e.g., QM/MM) level of theory by computing free energy corrections to classical results can be greatly facilitated through the use of force matching on classical generated ensembles. In particular, by adjusting only bonded type parameters (e.g., bond, angle, and dihedral terms), the configurational overlap of the classical model with the desired high level of theory is increased and thus improving the calculation of free energy corrections between levels of theory. Interaction Energy Hypothesis: Failure of the interaction energy hypothesis The need to achieve free energy calculations at highly accurate (e.g., QM/MM) levels of theory have resulted in various tricks and approximations to subvert the need for properly sampling conformationally relevant regions of the desired high level of theory. One such approach that has gained popularity is the so-called Interaction Energy Hypothesis'' (IEH). Initially grounded in rigorous implementation, the approach focuses on the substitution of total potential energy differences between classical MM and high-level QM in rigorous free energy simulation approaches (e.g, free energy perturbation and Bennett's acceptance ratio) with changes in environment-solute interactions between MM and QM/MM, which has substantially smaller fluctuations and therefore improving calculation convergence. Our results demonstrate that IEH is not only rigorously incorrect as a substitute for the change in total potential energy between high and low levels of theory but that IEH can systematically provide the wrong result. Specifically, using IEH on two different low levels of theory to obtain a density functional tight binding solvation free energy of bis-2-chloro-diethyl ether produced two almost identical results (e.g., differing only by 0.16 kcal/mol), but both deviating from rigorous calculations by about 1.5 kcal/mol. This result is particularly concerning as it deludes practitioner into a false confidence garnered by consistency in findings. MPID FF Efforts: Development of the multipole induced dipole forcefield One of the biggest pushes in the development of modern force fields is the need to incorporate polarizability effects. However, present efforts to incorporate polarization are often incompatible with refinement approaches (e.g., correcting a fixed charge model to a drude based force field) or are somewhat underwhelming in performance (in results and computational expediency). This motivated us to start work on the creation of our own polarizable force field, the Multipole Induced Dipole (MPID) force field. Based on the generation of higher-order multipole moments from heavy atoms of a molecule, initial parameters were derived by direct mapping of the CHARMM drude force field onto MPID. However, this need not (and quite possibly should not) be the case, as the best parameterization might deviate from the initial drude based setup. Thus, we are working towards generating a parameter set derived independently of the drude force field, with a dedicated MPID water model validated through various phases of bulk water in conjunction with the computation of solvation free energies. Other projects (without description due to annual report character count limits) Methods for Predicting Physicochemical Properties Evaluation of Enhanced Sampling Methods for Accelerating Peptide Insertion in Membranes Application of novel binding free energy calculation protocols to the host-guest blind challenges Reservoir pH replica exchange (R-pH-REM) for constant pH simulations Hybrid QM and MM method for pKa prediction in explicit solvent Using VMMS to explain the role of hydronium ions nearby acidic residues as an important reason behind the high apparent dielectric constant inside proteins
Showing the most recent 10 out of 63 publications
