mosbi

mosbi implements an automated multi-algorithm ensemble biclustering approach that computes robust ensemble biclusters on molecular omics data by integrating results from multiple biclustering algorithms through an error model-supported similarity network.

Key Features:

• Multi-algorithm ensemble biclustering: Combines results from multiple biclustering algorithms, including 11 available and established biclustering algorithms, to compute robust ensemble biclusters.
• Standardized interfaces: Provides standardized interfaces for biclustering results that can be easily extended to new biclustering algorithms.
• Similarity network construction: Computes similarity networks of biclusters and filters for overlaps using a custom error model.
• Community detection: Uses louvain modularity to extract bicluster communities from the similarity network, which are then converted to ensemble biclusters.
• Network visualization: Includes several network visualization methods to give an intuitive and scalable overview of bicluster communities.

Scientific Applications:

• Molecular signature identification: Identifies robust group and disease-specific molecular signatures across heterogeneous omics data by overcoming single algorithm specificities.
• Patient stratification: Supports advanced patient stratification and disease subtyping strategies through automated signature mining from biomedical data.
• Transcriptomics analysis: Identifies sample groups and corresponding feature sets from transcriptomics data.
• Proteomics analysis: Identifies sample groups and corresponding feature sets from proteomics data.
• Metabolomics analysis: Identifies sample groups and corresponding feature sets from metabolomics data.
• Biological hypothesis generation: Supports biological hypothesis generation through scalable network-based visualization of bicluster communities.

Methodology:

The method computes similarity networks of biclusters, filters for overlaps using a custom error model, applies louvain modularity to extract bicluster communities from the similarity network, and converts these communities to ensemble biclusters.