PAPER--accelerating parallel evaluations of ROCS.

Area of Science:

• Computational chemistry
• Molecular modeling
• High-performance computing

Background:

• Graphics Processing Units (GPUs) offer superior computational power compared to Central Processing Units (CPUs).
• Gaussian molecular shape overlay is a crucial technique in computational chemistry and drug discovery.
• Existing implementations may have performance limitations or inaccuracies.

Purpose of the Study:

• To design and implement an open-source Gaussian molecular shape overlay algorithm optimized for NVIDIA GPUs.
• To evaluate the performance and accuracy of the new GPU implementation against existing methods.
• To identify and address errors in common approximations used in molecular shape overlay algorithms.

Main Methods:

• Development of PAPER, an open-source software package utilizing NVIDIA GPU acceleration.
• Implementation of Gaussian molecular shape overlay algorithm on GPU architecture.
• Benchmarking PAPER against a CPU-based reference implementation and the commercial ROCS package.
• Analysis of approximation errors in standard molecular shape overlay algorithms.

Main Results:

• PAPER demonstrates one to two orders of magnitude speedup over CPU implementations.
• Speedups exceeding one order of magnitude were observed compared to the OpenEye ROCS package.
• The study identifies and quantifies errors introduced by approximations in common algorithms.

Conclusions:

• PAPER provides a highly efficient and accurate solution for Gaussian molecular shape overlay on GPUs.
• The open-source implementation accelerates molecular shape analysis, benefiting computational chemistry research.
• Understanding approximation errors is critical for reliable molecular modeling results.