GeauxDock: Accelerating Structure-Based Virtual Screening with Heterogeneous Computing.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Structure-based virtual screening is crucial for identifying drug candidates.
• Accurate and fast molecular docking engines are essential for effective virtual screening.
• Optimized docking codes are paramount for computational drug design.

Purpose of the Study:

• To describe the implementation, tuning, and performance of GeauxDock, a novel molecular docking program.
• To evaluate GeauxDock's efficiency on heterogeneous computing platforms.
• To demonstrate GeauxDock's potential to accelerate structure-based virtual screening.

Main Methods:

• GeauxDock utilizes a Monte Carlo algorithm with a novel scoring function combining physics-based, statistical, and knowledge-based potentials.
• The program is optimized for heterogeneous computing, including multi-core CPUs, Intel Xeon Phi, and NVIDIA GPUs.
• Performance tuning was conducted on the framework and docking kernel for serial and parallel execution.

Main Results:

• GeauxDock demonstrates near-ideal scaling on shared-memory multi-core CPUs.
• Intel Xeon Phi provides a 1.9x performance improvement over a dual 10-core Xeon CPU.
• NVIDIA GPUs, specifically GeForce GTX 980, achieve a speedup of up to 3.5x.

Conclusions:

• GeauxDock effectively leverages heterogeneous architectures to accelerate structure-based virtual screening.
• The program offers significant speedups, enhancing the efficiency of drug discovery pipelines.
• GeauxDock is open-sourced, promoting wider accessibility and application in computational drug design.