Efficient Large-Scale Virtual Screening Based on Heterogeneous Many-Core Supercomputing System.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• The increasing size of virtual drug databases necessitates efficient molecular docking tools.
• Existing tools face challenges in speed and accuracy, especially for large molecules.

Purpose of the Study:

• To develop an optimized molecular docking system, Vina@QNLM 2.0, for large-scale virtual screening.
• To improve computational efficiency and scoring capabilities for large molecules.
• To enhance support for applications like reverse target finding.

Main Methods:

• Developed Vina@QNLM 2.0 utilizing heterogeneous multicore architecture processors.
• Implemented an improved parallel strategy for enhanced performance.
• Optimized docking speed and scoring for molecules with molecular weight > 500.

Main Results:

• Vina@QNLM 2.0 achieves a 20-fold speedup in single docking processes compared to using logical processing units alone.
• Demonstrated robust scalability of 80.01% for reverse target finding tasks scaled to 122,401 kernel groups.
• Successfully completed a reverse target search for nine glycan molecules against 10,094 proteins in under one hour.

Conclusions:

• Vina@QNLM 2.0 offers a significant advancement in molecular docking efficiency and accuracy.
• The system effectively accelerates large-scale virtual screening and complex tasks like reverse target identification.
• This tool holds substantial promise for accelerating drug discovery pipelines.