GPU-Accelerated Flexible Molecular Docking.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Virtual screening is essential for computational drug discovery.
• Accurate and efficient molecular docking is required for virtual screening.
• MedusaDock is a flexible protein-small molecule docking platform.

Purpose of the Study:

• To accelerate the coarse docking phase of MedusaDock using graphics processing units (GPUs).
• To develop algorithms and software building blocks for GPU-accelerated molecular docking.
• To evaluate the performance and quality of GPU-accelerated MedusaDock.

Main Methods:

• Developed GPU-accelerated algorithms for the coarse docking phase of MedusaDock.
• Implemented software building blocks for efficient GPU utilization.
• Evaluated performance using single-GPU and multi-GPU configurations on a dataset of 3875 protein-ligand complexes.

Main Results:

• Achieved significant performance improvements in the coarse docking phase of MedusaDock through GPU acceleration.
• Demonstrated the effectiveness of the developed algorithms and data structures for GPU-based molecular docking.
• Comprehensive evaluation confirmed the quality and speed-up using GPU acceleration.

Conclusions:

• GPU acceleration is a viable strategy for enhancing the performance of molecular docking software.
• The developed techniques can be applied to other structure-based molecular docking tools.
• This work contributes to faster and more efficient computational drug discovery pipelines.