Accelerated CDOCKER with GPUs, Parallel Simulated Annealing, and Fast Fourier Transforms.

Area of Science:

• Computational Biology
• Molecular Modeling
• Drug Discovery

Background:

• Protein-ligand docking is crucial for drug discovery and understanding biomolecular interactions.
• Existing docking algorithms face challenges in computational speed and accuracy, particularly for flexible molecules.
• CHARMM's CDOCKER program is a widely used tool for molecular docking.

Purpose of the Study:

• To significantly enhance the throughput and accuracy of the CDOCKER docking algorithm using GPU acceleration.
• To adapt Fast Fourier Transform (FFT)-based docking for efficient ligand search in protein-ligand interactions.
• To accelerate the CDOCKER protocol through parallel molecular dynamics (MD) simulated annealing on GPUs.

Main Methods:

• Implementation of FFT-based docking for ligand translational and rotational space search on GPU-accelerated platforms.
• Adaptation of parallel MD simulated annealing for rigid and flexible receptor docking on GPUs.
• Integration of these GPU-accelerated methods into the CDOCKER algorithm within the CHARMM program.

Main Results:

• Achieved a 15,000-fold speedup in ligand search space exploration using FFT docking on GPUs.
• Demonstrated practical exhaustive search of ligand space for rigid docking, enabling better scoring function assessment.
• Obtained over 20-fold acceleration for the CDOCKER protocol using parallel MD simulated annealing on GPUs.
• Significantly improved redocking performance of CDOCKER compared to other popular docking programs on benchmark datasets.

Conclusions:

• The GPU-accelerated CDOCKER platform offers a highly competitive and efficient solution for both rigid and flexible receptor docking.
• The implemented FFT and parallel MD approaches substantially improve docking accuracy and speed.
• This work facilitates advancements in scoring function development and accelerates the drug discovery pipeline.