Accelerating pairwise statistical significance estimation for local alignment by harvesting GPU's power

Area of Science:

• Bioinformatics
• Computational Biology
• High-Performance Computing

Background:

• Pairwise statistical significance is crucial for identifying related sequences in bioinformatics.
• Current methods are computationally intensive, limiting performance and scalability.

Purpose of the Study:

• To accelerate pairwise statistical significance estimation for local sequence alignment using Graphics Processing Units (GPUs).
• To improve the performance and scalability of sequence comparison in bioinformatics.

Main Methods:

• Developed a GPU implementation for pairwise statistical significance estimation.
• Utilized a tile-based scheme for contiguous data access in GPU global memory.
• Extended parallelization for position-specific substitution matrices and dual-GPU utilization.

Main Results:

• Achieved end-to-end speedups of nearly 250x (single-GPU) and 370x (dual-GPU) compared to CPU implementations.
• Demonstrated high GPU occupancy through optimized data access and thread management.
• Successfully applied the method to position-specific substitution matrices for improved accuracy.

Conclusions:

• Leveraging modern GPU performance is an effective strategy for accelerating pairwise statistical significance estimation.
• The developed GPU implementation offers a significant performance improvement for local sequence alignment tasks.