GPU-Darwin

GPU-Darwin accelerates read overlapping for de novo assembly by GPU-accelerating D-SOFT filtration and GACT alignment to speed long-read overlap detection and alignment.

Key Features:

• Hardware Acceleration: Utilizes GPUs to accelerate overlap detection, achieving a 109x speedup versus an Intel Xeon CPU with 8 threads and a 24x speedup versus an IBM Power8 machine with 64 threads.
• Gap Scoring Models: Supports both linear and affine gap scoring models while maintaining high performance.
• Execution Modes: Can operate on CPUs or leverage GPU acceleration based on available hardware.
• D-SOFT Filtration: Filters the search space by counting non-overlapping bases in matching K-mers within a diagonal band (Diagonal-band based Seed Overlapping based Filtration Technique).
• GACT Alignment: Performs alignment of arbitrary-length reads using constant traceback memory (Genomic Alignment using Constant Traceback memory) to enable compute-intensive Smith-Waterman–style alignment efficiently.

Scientific Applications:

• OLC de novo assembly: Facilitates pairwise comparison of reads to identify overlaps in Overlap-Layout-Consensus based de novo assembly workflows.
• Long-read sequencing analysis: Suitable for state-of-the-art third-generation long DNA reads such as PacBio sequencing data.
• Long-read alignment: Enables fast Smith-Waterman alignment between long DNA reads for genomic research applications.

Methodology:

Incorporates two main components: D-SOFT (Diagonal-band based Seed Overlapping based Filtration Technique), which filters by counting non-overlapping bases in matching K-mers within a diagonal band, and GACT (Genomic Alignment using Constant Traceback memory), which aligns reads of arbitrary length using constant traceback memory for compute-intensive steps.