GHOSTM: a GPU-accelerated homology search tool for metagenomics

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Metagenomic analysis requires sensitive homology searches to map DNA fragments to protein sequences.
• Current tools like BLAST lack the speed for next-generation sequencing data, while faster tools like BLAT lack sensitivity.
• There is a critical need for efficient and sensitive homology search tools in metagenomics.

Purpose of the Study:

• To develop a highly efficient homology search algorithm optimized for graphics processing unit (GPU) calculations.
• To implement this algorithm as a GPU system named GHOSTM.
• To address the computational challenges posed by the increasing volume of sequence data in metagenomics.

Main Methods:

• Developed a novel homology search algorithm leveraging GPU acceleration.
• Implemented a two-step process: candidate position searching using pre-calculated indexes and local alignment calculation.
• Executed both indexing and alignment processes on GPUs.

Main Results:

• GHOSTM achieved 130x and 407x speed improvements over BLAST using 1 and 4 GPUs, respectively.
• GHOSTM demonstrated superior search sensitivity compared to BLAT.
• GHOSTM was 4x and 15x faster than BLAT with 1 and 4 GPUs, respectively.

Conclusions:

• GHOSTM is a GPU-optimized algorithm for sensitive sequence homology searches.
• The system provides a cost-efficient solution for analyzing the exponentially growing volume of sequence data.
• GHOSTM addresses the limitations of current tools in handling large-scale metagenomic datasets.