BLAMM: BLAS-based algorithm for finding position weight matrix occurrences in DNA sequences on CPUs and GPUs.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying matches of position weight matrices (PWMs) in large DNA sequences is computationally intensive.
• Existing algorithms are efficient but complex, posing implementation challenges.

Purpose of the Study:

• To introduce BLAMM, a novel, simple, and efficient tool for large-scale PWM matching.
• To leverage high-performance computing techniques for accelerated DNA sequence analysis.

Main Methods:

• Expressing the workload as matrix-matrix products for optimized evaluation using BLAS libraries.
• Implementing a parallelizable algorithm suitable for both CPUs and GPUs.
• Developing a tool with runtime independent of the chosen p-value.

Main Results:

• BLAMM demonstrates competitive single-core performance against state-of-the-art PWM matching software.
• Significant efficiency gains are achieved with multithreading, outperforming existing tools.
• The tool requires negligible memory, enabling large-scale analyses like scanning the entire human genome.
• Scanning the human genome for 1404 PWMs takes 13 minutes on a 36-core machine and under 5 minutes on a dual GPU system.

Conclusions:

• BLAMM provides an efficient and accessible solution for identifying PWM matches in extensive DNA datasets.
• The C++ source code is publicly available, promoting further research and application.