jNMFMA

jNMFMA identifies differentially expressed genes across heterogeneous omics datasets by applying joint non-negative matrix factorization to capture dependence structures in transcriptomics data.

Key Features:

• Joint Non-Negative Matrix Factorization (jNMF): Extends non-negative matrix factorization (NMF) to jointly decompose multiple transcriptomics data matrices into one common submatrix and several individual submatrices.
• Metagene-based dimensionality reduction: Maps high-dimensional transcriptomics data into a lower-dimensional space defined by metagenes that represent underlying biological signals.
• DEG identification via metagenes: Identifies differentially expressed genes as genes associated with differentially expressed metagenes.
• Multi-omics compatibility: Detects DEGs across different omics types including gene expression and DNA methylation.
• Performance validation: Demonstrated effectiveness on simulated and real-world cancer datasets with superior performance compared to other approaches.

Scientific Applications:

• Meta-analysis of heterogeneous omics data: Joint analysis of multiple transcriptomics datasets to improve robustness of DEG discovery.
• Integrative analysis of gene expression and DNA methylation: Identification of DEGs showing consistency across gene expression and DNA methylation data.
• Discovery of latent biological signals: Extraction of metagenes that capture dependence structures and underlying biology in high-dimensional datasets.
• Benchmarking on cancer datasets: Evaluation and comparative assessment using simulated and real-world cancer datasets.

Methodology:

Extends NMF to a joint NMF framework and simultaneously decomposes multiple transcriptomics matrices into a common submatrix and individual submatrices; projects data into a metagene-defined lower-dimensional space and identifies DEGs as genes associated with differentially expressed metagenes.