Dr. David Cerutti
Center for Structural Biology
Vanderbilt University
Nashville, TN

Monday, February 23, 2009
3:00 pm (CT)
3269 Beckman Institute

Abstract

To reach reasonable conclusions, simulations of any complex system must satisfy two basic challenges, accuracy of the cost function and adequacy of sampling. Biomolecular simulations require validation of the underlying force fields and better algorithms compatible with massively parallel computing to carry out simulations on biologically relevant time scales. To meet the cost function requirement, we have taken a comprehensive approach to modeling and simulating protein crystal lattices in atomic detail, including crystallization agents and cryoprotectants present in the X-ray diffraction experiments. Comparison of these results to more conventional solution-phase simulations provides a better link between theory and experiment, improving our ability to explain certain aspects of the solution phase simulations and the crystal structures themselves. This approach also illuminates targets for improvement in the molecular force fields; we are pursuing simulations to test an array of popular force fields and water models against a protein crystal structure resolved to 0.96A at room temperature. To address the sampling requirement, we propose extensions of the Particle Mesh Ewald (PME) method commonly used to compute electrostatics in simulations of thousands of atoms. Whereas traditional PME involves a convolution of a relatively fine charge density mesh to obtain the electrostatic potential, the "Staggered Mesh Ewald" (StME) method involves convolutions of two coarse charge density meshes, allowing up to 40% reduction in the amount of data that must be communicated without sacrificing accuracy in the computed forces. Some of the principles behind the success of StME can also be applied to traditional PME implementations for improved accuracy at negligible cost. We will conclude with a brief overview of the next generation of biomolecular force fields and their implementations for increasingly detailed simulations.