Poisson-Boltzmann equation (PBE)-based implicit solvent models have been widely used in computational studies of biomolecules. The main difficulty for these models is how to set up the discretized equation near the solute-solvent interface. We have studied and implemented several new methods to improve its performance, including the second-order harmonic averaging method and the -factor harmonic averaging method. We specifically designed these new algorithms to have uniform banded structure for their discretized equations, so they would be more suitable for parallel computing and thus gain a tremendous efficiency on Graphical Processing Units (GPUs). Furthermore, we also developed a new method for computing the solute-solvent interface on the GPUs, because the solvent excluded surface is highly inefficient as it is a serial algorithm based on geometry consideration. The new method was obtained in a machine learned way, and it is naturally suitable for parallel computing on GPUs. Along with other developments on supporting functions, we expect the current works lay the foundation for highly parallel efficient PBE modeling of biomolecular systems on GPUs.

Improving Poisson-Boltzmann modeling for parallel computing on graphical processing units

ACS Fall 2021

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