GPU acceleration of Dock6's Amber scoring computation.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Bioinformatics

Background:

• Virtual screening is crucial for accelerating drug discovery, but scoring methods are computationally intensive.
• Existing scoring functions, like Amber scoring in Dock6, require extensive floating-point calculations, often taking weeks or months.
• The need for rapid analysis is critical during outbreaks of infectious diseases (e.g., SARS, H1N1).

Purpose of the Study:

• To accelerate the Amber scoring functionality within the Dock6 software using GPU computation.
• To investigate performance-influencing factors when porting Amber scoring to a GPU platform.
• To enhance the efficiency and competitiveness of virtual screening for large datasets.

Main Methods:

• Leveraged NVIDIA CUDA (Compute Unified Device Architecture) platform to port Amber scoring to GPUs.
• Analyzed and optimized factors such as thread management, data transfer, and hidden divergence for GPU performance.
• Compared the performance of the GPU-accelerated version against the original CPU-based implementation.

Main Results:

• Achieved a 6.5x speedup for Amber scoring using GPU acceleration compared to a dual-core CPU.
• Demonstrated the feasibility and significant performance gains of GPU computing for docking scoring.
• Identified key optimization strategies for GPU porting of computational chemistry tasks.

Conclusions:

• GPU-accelerated Amber scoring significantly reduces computational time for virtual screening.
• This approach makes large-scale virtual screening more practical and efficient.
• The optimized GPU implementation enhances the utility of Dock6 in drug discovery pipelines.