Bwa-mem2

Bwa-mem2 aligns short DNA reads to reference genomes, accelerating the BWA-MEM algorithm for high-throughput NGS data mapping.

Key Features:

• Output compatibility: Produces alignment outputs identical to BWA-MEM.
• Performance improvement: Offers approximately 1.3–3.1× faster runtime than BWA-MEM depending on dataset size and hardware configuration.
• Architecture-aware implementation: Optimized for large-scale data and multicore processors in clusters and cloud environments.
• Kernel focus: Targets three main computational kernels that account for over 85% of total compute time.
• Optimization techniques: Implements enhanced cache reuse, algorithm simplification, consolidation of memory allocations, software-based data prefetching, and utilization of SIMD instructions.
• Code reorganization: Substantial source-code restructuring to enable low-level and architecture-specific optimizations.
• Measured speedups: Achieved nearly 2×, 183×, and 8× speedups on the three critical kernels and up to 3.5× (single thread) and 2.4× (single socket) end-to-end improvements on an Intel Xeon Skylake processor.

Scientific Applications:

• Short-read mapping: Aligns short DNA reads to reference genomes for downstream genomic analyses.
• Variant calling pipelines: Serves as the mapping component for workflows such as the Genome Analysis Toolkit (GATK).
• High-throughput sequencing processing: Handles large outputs from NGS platforms such as the Illumina NovaSeq 6000.
• Scalable computation: Enables mapping at scale on clusters and cloud-based systems using multicore processors.

Methodology:

Uses an architecture-aware implementation with kernel-level optimization of three primary compute kernels via enhanced cache reuse, algorithm simplification, consolidated memory allocations, software-based data prefetching, SIMD instructions, and extensive source-code reorganization.