Task-parallel message passing interface implementation of Autodock4 for docking of very large databases of compounds...

Area of Science:

• Computational chemistry
• Bioinformatics
• Molecular modeling

Background:

• Autodock4 is a widely used grid-based molecular docking program.
• The serial version of Autodock4 has limitations in processing large compound libraries efficiently.
• Accelerating molecular docking is crucial for high-throughput drug discovery.

Purpose of the Study:

• To develop and evaluate a Message Passing Interface (MPI)-based parallel implementation of Autodock4.
• To improve the computational efficiency and scalability of the Autodock4 docking program.
• To enable simultaneous and independent docking of multiple compounds on large-scale computing clusters.

Main Methods:

• Developed Autodock4.lga.MPI, a parallel version utilizing the Lamarkian genetic algorithm.
• Implemented efficient handling of precalculated grid files in a single binary format.
• Optimized input/output operations and MPI communication for performance.
• Investigated optimal docking strategies based on ligand flexibility and energy evaluations.

Main Results:

• Autodock4.lga.MPI demonstrates significant speedup, scaling effectively up to 8192 CPUs with minimal overhead.
• The MPI version drastically reduces input/output activity compared to the serial version.
• Achieved optimal docking performance by ordering ligands by flexibility and adjusting energy evaluations.
• In 24 hours, the system can dock approximately 300,000 small flexible compounds or 11 million rigid compounds.

Conclusions:

• The MPI-based implementation of Autodock4 offers a highly scalable and efficient solution for large-scale molecular docking.
• This advancement significantly accelerates the drug discovery process by enabling rapid screening of vast compound libraries.
• The optimized docking strategy further enhances the utility of this parallel computing approach.