MoSBi

MoSBi identifies coordinated sample groups and associated molecular feature sets within heterogeneous omics datasets using biclustering to support molecular patient stratification and disease subtyping.

Key Features:

• Unsupervised learning: Leverages unsupervised machine learning techniques, specifically biclustering, for pattern discovery in omics data.
• Biclustering: Simultaneously identifies groups of samples and their corresponding feature sets across diverse molecular data.
• Multi-algorithm ensemble: Automatically integrates results from multiple biclustering algorithms to enhance robustness.
• Error model-supported similarity network: Merges biclustering outputs using an error model-supported similarity network to mitigate algorithm-specific biases and parameter sensitivities.
• Cross-omics evaluation: Performance has been evaluated across transcriptomics, proteomics, metabolomics, and synthetic datasets with diverse properties.
• Network-based visualization: Provides scalable network representations of bicluster communities to support interpretation and hypothesis generation.

Scientific Applications:

• Patient stratification: Identifies molecularly defined patient subgroups for stratified analyses.
• Disease subtyping: Discerns disease-specific subtypes based on coordinated molecular signatures.
• Discovery of molecular signatures: Detects robust disease-associated and condition-specific feature sets across omics layers.
• Hypothesis generation: Uses bicluster communities and network representations to generate biological hypotheses about molecular mechanisms.

Methodology:

Applies unsupervised biclustering algorithms in an automated multi-algorithm ensemble and integrates results via an error model-supported similarity network, with evaluation performed on transcriptomics, proteomics, metabolomics, and synthetic datasets.