GPU-Acceleration of Sequence Homology Searches with Database Subsequence Clustering

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Sequence homology searches are computationally intensive, particularly for large-scale metagenomic datasets.
• Graphics processing units (GPUs) offer a cost-effective solution for high-performance computing in biological analyses.

Purpose of the Study:

• To accelerate the GHOSTZ protein sequence homology search algorithm by leveraging GPU parallel processing.
• To optimize memory access for both GPU computations and CPU-GPU communication.

Main Methods:

• The core computational steps of the GHOSTZ algorithm were mapped onto a GPU, creating the GHOSTZ-GPU implementation.
• Memory access patterns were optimized for efficient GPU calculations and data transfer between the CPU and GPU.

Main Results:

• GHOSTZ-GPU with one GPU and 12 CPU threads achieved a 3.0- to 4.1-fold speedup compared to GHOSTZ with 12 CPU threads.
• Utilizing three GPUs alongside 12 CPU threads resulted in an approximate 5.8- to 7.7-fold increase in speed over the CPU-only version.

Conclusions:

• GHOSTZ-GPU provides a significant acceleration for protein sequence homology searches, especially beneficial for metagenomic applications.
• GPU acceleration is a viable strategy to overcome the computational bottlenecks in large-scale sequence analysis.