Recently the rapid advance of GPU-accelerated computing has made significant impact on scientific, analytics, engineering, consumer, and enterprise applications. In the area of molecular dynamics (MD) based alchemical free energy simulations, currently the most accurate approaches for ligand binding problems, GPU-accelerated implementation has caught broad attention only very recently. Here we report the advancement of the GPU-accelerated pmemdGTI package for alchemical free energy methods. This package has been integrated into the official AMBER 18 release, and is fully supported with the full spectrum of features in AMBER. Additionally, the package enables full functionality to several new features that are in the AMBER pipeline, but that will not part of the next official release, including new enhanced sampling methods, metal ion potentials and integration with new Parameter-Interpolated Thermodynamic Integration (PI-TI) methods. Our implementation enables speed-ups of over two orders of magnitude with the fastest GPU (e.g., V100) compared to a single CPU core. As a result, the GPU hardware is able to bring to the desktop the computing power for free energy simulations that once only was achievable through expensive supercomputers with many compute cores. The pmemdGTI code is also available as binary executables for Windows desktop machines. We demonstrate accurate, reliable, and efficient verification and prediction of ligand binding potency results for selected biological systems, including metal regulated ligand binding. Results also demonstrate the method is a powerful predictive tool for drug discovery able to provide atomic-level insight into the origins of drug-resistance.

GPU-accelerated alchemical free energy simulations: A new tool for accurate, affordable prediction of ligand binding potency for drug discovery, optimization, and resistance problems

255th National Meeting (2018)

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