PatternMarkers & GWCoGAPS for novel data-driven biomarkers via whole transcriptome NMF.

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Non-negative Matrix Factorization (NMF) is used to associate gene expression with biological processes.
• Existing NMF methods can obscure important gene biomarkers.
• Novel methods are needed for unbiased biomarker discovery from whole-genome data.

Purpose of the Study:

• To develop a novel statistic, patternMarkers, for extracting genes for biological validation and visualization of NMF results.
• To develop Genome-Wide CoGAPS Analysis in Parallel Sets (GWCoGAPS), a robust whole-genome Bayesian NMF method.
• To provide tools for data-driven biomarker discovery from whole-genome expression data.

Main Methods:

• Developed the patternMarkers statistic for NMF analysis.
• Developed Genome-Wide CoGAPS Analysis in Parallel Sets (GWCoGAPS), a Bayesian NMF algorithm using MCMC.
• Created a manual version of GWCoGAPS with analytic and visualization tools, including the patternMatcher Shiny web application.

Main Results:

• GWCoGAPS is the first robust whole-genome Bayesian NMF method.
• The patternMatcher tool facilitates visualization and analysis of NMF results.
• Applied GWCoGAPS and patternMarkers to GTEx data, identifying granular brain-region and cell-type specific signatures.

Conclusions:

• GWCoGAPS and patternMarkers enable the ascertainment of data-driven biomarkers from whole-genome data.
• These tools enhance the biological validation and visualization of NMF results.
• The developed methods successfully identified specific gene signatures and biomarkers in human brain data.