ICA

ICA decomposes omics and other biomedical datasets into statistically independent components to identify underlying biological signals and sources in cancer research.

Key Features:

• Matrix factorization / Blind source separation: Performs matrix factorization framed as blind source separation to recover independent signals from mixed measurements.
• Dimensionality Reduction: Reduces high-dimensional omics data while preserving biologically relevant variation for downstream analysis.
• Deconvolution: Separates mixed cellular or molecular signals within bulk and single-cell omics datasets to identify distinct biological components.
• Data Pre-processing and Meta-analysis: Supports preprocessing of omics matrices and aggregation across studies for meta-analysis.
• Optimal Component Determination (Stability Analysis): Employs stability analysis to estimate the optimal number of independent components for robust decomposition.
• Reproducibility and Comparison: Provides a framework for assessing reproducibility of components and comparing ICA results with PCA and NMF.
• Integrative Multi-level Omics Analysis: Integrates transcriptome, methylome, proteome, and single-cell data to reveal functional subsystems and interactions.
• BIODICA pipeline: Example implementation that computes independent components from large-scale omics data and applies stability analysis for component selection.

Scientific Applications:

• Tumor omics deconvolution: Disentangles cellular and molecular mixtures in tumor transcriptome, methylome, proteome, and single-cell datasets.
• Feature extraction for downstream analyses: Extracts independent components used as features for classification, clustering, or network analysis.
• Cross-study meta-analysis: Harmonizes components across studies to identify reproducible biological signals.
• Identification of functional subsystems: Reveals subsystems and condition-specific interactions within complex biological systems.
• Biomedical signal separation: Applies to non-omics biomedical data such as fMRI for source separation tasks.
• Method benchmarking: Facilitates comparison of ICA performance against PCA and NMF in matrix factorization tasks.

Methodology:

Independent Component Analysis implemented as matrix factorization / blind source separation, stability analysis to select the optimal number of components, and application to dimensionality reduction, deconvolution, data pre-processing, meta-analysis, and comparison with PCA and NMF (implemented e.g. in BIODICA).