Hierarchical parallelization of gene differential association analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray gene differential expression and association analyses are computationally intensive due to high-dimensional data.
• Leveraging multicore computing is crucial for accelerating these demanding analyses.
• Existing methods often struggle with the computational load of permutation-based procedures.

Purpose of the Study:

• To present a novel two-layer hierarchical parallel implementation for gene differential association analysis.
• To effectively utilize both fine- and coarse-grain parallelism on modern multicore systems.
• To address the computational demands of large-scale biological data analysis.

Main Methods:

• Developed a two-layer hierarchical parallel implementation strategy.
• Integrated fine- and coarse-grain parallelism based on communication frequency.
• Optimized data sharing to match underlying hardware properties on parallel processing systems.

Main Results:

• The hierarchical strategy reduced memory and bandwidth requirements.
• Achieved significant reductions in computation time for gene differential association analysis.
• Demonstrated effective scaling of the analysis code with an increasing number of processors.

Conclusions:

• The optimal performance is achieved when thread working sets fit within machine cache.
• Recommends aligning the number of threads per MPI process with cluster cache capabilities.
• The hierarchical parallelization principles are applicable to other computationally intensive tasks.