ParBiBit

ParBiBit performs parallel biclustering of binary datasets to identify two-dimensional submatrices of correlated rows and columns for applications including gene expression analysis, marketing, and text mining.

Key Features:

• Parallel Processing: Implemented in C++11 with multi-threading and MPI to run on distributed-memory systems composed of multicore CPU nodes.
• Efficiency and Performance: Demonstrated substantial performance improvements relative to BiBit in evaluations on 18 representative datasets across two eight-node systems.
• Methodology Consistency: Retains the underlying BiBit methodology for identifying biclusters by grouping binary information into patterns.
• Application Scope: Targets binary datasets commonly found in gene expression analysis and extends to other domains such as marketing and text mining.

Scientific Applications:

• Gene expression analysis: Accelerates identification of correlated gene expression patterns represented as binary matrices.
• Genetics: Supports detection of two-dimensional associations among genetic markers or samples encoded as binary data.
• Marketing: Enables discovery of co-occurring binary customer behaviors or attributes.
• Text mining: Facilitates finding biclusters in binary document-term matrices to reveal co-occurring terms and document groups.

Methodology:

ParBiBit applies the BiBit biclustering methodology by grouping binary entries into patterns to identify biclusters and parallelizes computation using C++11, multi-threading and MPI on distributed-memory multicore CPU nodes.