A key advantage of the 3D reference interaction site model (3D-RISM) theory of molecular solvation for biomolecular systems is its ability to quantitatively model solvent density distributions, forces, and thermodynamics using explicit solvent models. However, 3D-RISM remains much more computationally expensive than other implicit solvent models, such as generalized-Born or Poisson-Boltzmann. To accelerate 3D-RISM, we have targeted the calculation of the long-range electrostatic interactions and convergence of the density distribution. For all long-range electrostatic interactions, including long-range asymptotics and forces, we have implemented a cluster-particle treecode approach, which offers O(N logN) scaling when dealing with N solute charges. This provides significant improvements over the direct-sum scaling of O(N2). Convergence of the density distribution currently scales as O(N logN) due to the modified direct inversion of the iterative subspace (MDIIS) but is still computationally costly on typical grid sizes. By coupling MDIIS with a multigrid approach, we reduce the number of fine grid iterations by 50 to 90%. Together, these methods provide O(N logN) scaling for 3D-RISM and accelerate solutions by a factor of 10 or more for large proteins.

Accelerating the 3D-RISM implicit solvent model using treecode and multigrid methods

256th National Meeting (2018)

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