BEM

BEM performs Boolean Matrix Factorization using Expectation Maximization to extract biologically meaningful bicluster structures from transcriptomic datasets.

Key Features:

• Alignment with molecular mechanisms: Models binary coregulation patterns that reflect gene regulatory processes rather than relying on linear combinations.
• Expectation Maximization: Uses expectation maximization to iteratively refine factorization parameters.
• Scalability: Handles matrices exceeding 100 million data points for large-scale analyses.
• Bicluster structure extraction: Extracts clear bicluster structures from transcriptomic data to capture coregulation patterns.
• Versatility across data types: Applicable to bulk RNA-seq, single-cell RNA-seq, and spatial transcriptomics datasets.
• Performance (reconstruction error): Synthetic experiments report lower reconstruction error compared with other Boolean matrix factorization methods.
• Disease and cell-type correlation: With appropriate binarization, identifies coregulation patterns consistent with disease subtypes, cell types, or spatial anatomy.

Scientific Applications:

• Tumor signal perturbation and subtype classification: Reveals tumor signal perturbation status and supports subtype classification from transcriptomic data.
• Coregulation pattern discovery: Uncovers coregulation patterns that inform molecular mechanisms of disease.
• Cross-platform transcriptomic analysis: Facilitates analysis of bulk RNA-seq, single-cell RNA-seq, and spatial transcriptomics for integrative studies.
• Support for precision medicine research: Provides coregulation-derived signals that can inform personalized medicine and targeted therapy investigations.

Methodology:

BEM performs Boolean Matrix Factorization using Expectation Maximization for iterative parameter refinement, applies binarization to detect coregulation patterns, and is evaluated using reconstruction error in synthetic experiments.