Bi-Force

Bi-Force applies a weighted bi-cluster editing model to perform biclustering and uncover local patterns in high-dimensional biological datasets.

Key Features:

• Weighted bi-cluster editing model: Uses a weighted bi-cluster editing formalism to identify biclusters representing local patterns in data.
• Combinatorial optimization on graphs: Models biclustering as a combinatorial optimization problem and employs graph-based techniques for analysis.
• Flexibility across data types: Processes any set of biological entities given pairwise similarity measures and handles diverse multi-condition datasets beyond traditional gene expression.
• Implementation and integration: Implemented in Java and integrated into the BiCluE software package.
• Benchmarking against established methods: Evaluated in comparison to FABIA, QUBIC, Cheng and Church, Plaid, BiMax, Spectral, xMOTIFs, and ISA following the Eren et al. (2013) evaluation protocol.
• Dataset testing: Tested on synthetic datasets and nine large gene expression datasets from the Gene Expression Omnibus (GEO).
• Functional validation: Resulting biclusters were validated using Gene Ontology enrichment analysis.

Scientific Applications:

• Local pattern discovery in gene expression: Identifies coherent subsets of genes and conditions in gene expression datasets from GEO.
• Analysis of multi-condition datasets: Detects local patterns across diverse biomedical and multi-condition datasets using pairwise similarity measures.
• Functional interpretation: Produces biclusters suitable for downstream Gene Ontology enrichment to assess biological relevance.

Methodology:

Uses a weighted bi-cluster editing model and combinatorial optimization on graphs; evaluated on synthetic and nine GEO gene expression datasets, validated by Gene Ontology enrichment, and compared using the Eren et al. (2013) evaluation protocol.