A high-performance parallel-generalized Born implementation enabled by tabulated interaction rescaling.

Area of Science:

• Computational Chemistry
• Molecular Dynamics Simulations

Background:

• Implicit solvent models, particularly Generalized Born (GB), reduce system complexity for efficient simulations.
• Challenges remain in GB implementations outperforming explicit solvent models due to complex calculations.

Purpose of the Study:

• To develop a computationally efficient Generalized Born method for molecular simulations.
• To enhance the performance of implicit solvent simulations on modern hardware.

Main Methods:

• A rescaling transformation was applied to the Generalized Born expression, simplifying it to a single variable.
• Tabulated implementation and single-instruction multiple-data (SIMD) acceleration were utilized in Gromacs.
• Adaptation for virtual interaction sites and streaming kernels without neighborlists were implemented.

Main Results:

• The new method achieves performance within a factor of 2 of in vacuo simulations.
• A simulation performance of 0.86 micros/day was reached for BBA5 with a 1-nm cutoff on a desktop processor.
• Optimized implementations accelerate both cutoff and non-cutoff based implicit solvent simulations.

Conclusions:

• The developed Generalized Born method significantly enhances computational efficiency for molecular dynamics.
• This approach makes implicit solvent simulations more practical for high-throughput studies and complex systems.