ParBiBit: Parallel tool for binary biclustering on modern distributed-memory systems.

Area of Science:

• Computer Science
• Bioinformatics
• Data Mining

Background:

• Biclustering identifies correlated row and column submatrices in large datasets.
• Binary datasets are prevalent in genetics, marketing, and text mining.
• Existing tools like BiBit offer accuracy but can be slow for large-scale analysis.

Purpose of the Study:

• To develop ParBiBit, a parallel tool for accelerating bicluster search in binary datasets.
• To leverage modern distributed-memory systems for enhanced computational performance.
• To maintain the accuracy of the established BiBit sequential algorithm.

Main Methods:

• ParBiBit is a parallel implementation of the BiBit sequential algorithm, utilizing C++11.
• The tool employs threads and Message Passing Interface (MPI) processes for distributed computing.
• It groups binary data into patterns, mirroring BiBit's methodology.

Main Results:

• ParBiBit demonstrates significant performance improvements over the sequential BiBit tool.
• Evaluations on 18 datasets across two eight-node systems confirm its speed advantage.
• The tool provides identical results to BiBit, ensuring comparable accuracy.

Conclusions:

• ParBiBit offers a scalable and efficient solution for bicluster analysis on large binary datasets.
• The parallel implementation effectively utilizes multicore CPUs and distributed systems.
• Open-source code and documentation are available for broader adoption and research.