Toward Fast and Accurate Binding Affinity Prediction with pmemdGTI: An Efficient Implementation of GPU-Accelerated...

Area of Science:

• Computational chemistry
• Molecular dynamics simulations
• Drug discovery

Background:

• Calculating ligand-protein binding affinities is crucial for drug discovery.
• Existing computational methods can be time-consuming, limiting throughput.
• GPU acceleration offers potential for significant speed-ups in molecular simulations.

Purpose of the Study:

• To implement and evaluate a GPU-accelerated thermodynamic integration method for calculating binding affinities.
• To assess the performance and accuracy of the new code compared to CPU-based methods.

Main Methods:

• Implementation of the thermodynamic integration method within the pmemd module of AMBER 16.
• Utilizing Graphics Processing Units (GPUs) for accelerated computation (pmemdGTI).
• Comparison of computational speed and numerical accuracy against single CPU core calculations.

Main Results:

• The pmemdGTI code achieves over a 2-order of magnitude speed-up compared to a single CPU core.
• No statistically significant numerical differences were observed between GPU and CPU calculations.
• Demonstrated feasibility and efficiency of GPU acceleration for binding affinity calculations.

Conclusions:

• The pmemdGTI code provides a powerful and efficient new tool for drug discovery.
• GPU acceleration significantly enhances the speed of calculating ligand-protein binding affinities.
• The method offers accurate results, enabling faster lead optimization in drug development.