Parallel implementation of 3D protein structure similarity searches using a GPU and the CUDA

Area of Science:

• Structural bioinformatics
• Computational biology
• Biophysics

Background:

• 3D protein structure similarity searching is crucial in structural bioinformatics.
• Current methods face computational complexity challenges, especially with large databases.
• Efficient algorithms are needed to accelerate the identification of common molecular substructures.

Purpose of the Study:

• To present a GPU-based implementation of the CASSERT algorithm for accelerated 3D protein structure similarity searching.
• To demonstrate the efficiency gains of parallelizing structure alignment phases on GPUs.
• To enable real-time 3D structure similarity searches.

Main Methods:

• Implementation of the CASSERT algorithm on a graphics processing unit (GPU).
• Two-phase alignment strategy for matching protein structure fragments.
• Massive parallelization of alignment processes across GPU cores.

Main Results:

• The GPU-CASSERT implementation achieved an average 180-fold speed increase compared to the CPU version.
• Significant reduction in execution time for 3D structure similarity searches.
• Demonstrated feasibility of real-time analysis for large-scale structure comparisons.

Conclusions:

• Massive parallelization on GPU devices drastically reduces 3D structure similarity search times.
• GPU-CASSERT enables efficient and rapid searching within extensive protein structure databases.
• The developed method facilitates real-time structural bioinformatics analyses.