PJD

PJD performs partial and joint decomposition of biologically structured gene expression matrices using low-rank models to enable integrative analysis across multiple genetic, transcriptomic, and epigenetic data sets.

Key Features:

• Low-rank modeling: Implements low-rank models to represent biologically structured gene expression matrices.
• Decomposition styles: Provides four decomposition styles: separately, concatenately, jointly, and statistically.
• Two-stage linked component analysis (2s-LCA): Implements 2s-LCA for joint decomposition of multiple biologically related experimental data sets.
• Data integration: Integrates genetic, transcriptomic, and epigenetic data collections in the decomposition.
• Variation and noise handling: Models and separates biological variation, unwanted noise, varying target measurements, and batch effects.
• Visualization and analysis: Facilitates visualization and analysis of gene expression matrices through low-rank decompositions.
• Evaluation: Reports consistency and empirical performance assessed via simulation studies.

Scientific Applications:

• Integrative analysis: Enables integrative analysis across multiple biologically related experimental data sets.
• Pattern discovery: Dissects and interprets complex patterns in multi-dimensional gene expression data.
• Batch effect mitigation: Identifies shared biological processes obscured by batch effects, varying measurements, and noise.
• Neurodevelopment analysis: Applied to four human brain development data sets to reveal shared structural gene expression patterns during neurogenesis.

Methodology:

PJD applies low-rank matrix decompositions in four modes (separately, concatenately, jointly, statistically) and implements two-stage linked component analysis (2s-LCA); methods were evaluated by simulation studies.