Accelerating two algorithms for large-scale compound selection on GPUs.

Area of Science:

• Computational Chemistry
• Drug Discovery Informatics
• Bioinformatics and Cheminformatics

Background:

• Molecular similarity and diversity are crucial for effective compound selection in drug discovery.
• Existing selection algorithms can be computationally intensive, especially for large compound datasets.
• The need for faster computational methods is critical to accelerate the drug discovery pipeline.

Purpose of the Study:

• To accelerate two widely used compound selection algorithms, the leader and spread algorithms.
• To implement these algorithms on graphical processing units (GPUs) for enhanced performance.
• To evaluate the speedup achieved by GPU acceleration compared to traditional CPU implementations.

Main Methods:

• Parallelization of molecular similarity calculations using Daylight fingerprints and the Tanimoto index.
• Implementation of the leader and spread algorithms on GPU hardware utilizing the open-source Thrust library.
• Performance comparison between GPU-accelerated and CPU-based algorithm execution.

Main Results:

• The GPU-accelerated leader algorithm demonstrated a speedup of 73-120 times over its CPU counterpart.
• The GPU-accelerated spread algorithm achieved a speedup of 78-143 times compared to the CPU version.
• Significant performance gains were observed for both algorithms when executed on GPUs.

Conclusions:

• GPU acceleration offers a substantial improvement in the computational efficiency of compound selection algorithms.
• This approach can significantly reduce the time required for analyzing large chemical libraries in drug discovery.
• The optimized algorithms are valuable tools for accelerating lead identification and optimization processes.