BLVector

BLVector accelerates protein sequence alignment by implementing a vectorized Smith-Waterman algorithm with affine gap penalties and substitution matrices (BLOSUM62, PAM30) optimized for x86 CPUs with AVX-512 to exploit multicore vectorization.

Key Features:

• Vectorized Smith-Waterman: Implements a vectorized Smith-Waterman algorithm for local protein alignments.
• Affine gap penalties: Supports affine gap penalty scoring for improved alignment accuracy.
• Substitution matrices: Uses score matrices such as BLOSUM62 and PAM30 for protein scoring.
• AVX-512 optimization: Optimized for x86 architectures with AVX-512 to leverage SIMD vectorization.
• Multicore exploitation: Exploits multicore CPU capabilities to improve throughput.
• Performance on mid-size proteins: Demonstrates a 3.2-fold reduction in execution time versus BLAST+ for mid-size proteins (~750 amino acids).
• Result retrieval trade-off: May require more time than BLAST+ for longer proteins but retrieves a substantially larger set of results.
• Heuristic approach: Operates as a fine-tuned heuristic similar to BLAST+, prioritizing speed over exhaustive search and producing differences primarily in low-score alignments.

Scientific Applications:

• Mid-size protein alignment: High-throughput alignment of mid-size protein sequences (~750 amino acids) with improved speed.
• Large dataset processing: Rapid processing of large protein sequence datasets where time efficiency is critical.
• Complementary analysis to BLAST+: Serves as a complementary option to BLAST+ when users require different trade-offs between speed and breadth of retrieved results.

Methodology:

Vectorized implementation of the Smith-Waterman algorithm with affine gap penalties using substitution matrices (BLOSUM62, PAM30), optimized for x86 AVX-512 and multicore execution, implemented as a fine-tuned heuristic comparable to BLAST+.