Gene Ranking of RNA-Seq Data via Discriminant Non-Negative Matrix Factorization

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA sequencing (RNA-seq) is crucial for transcriptome analysis, but high dimensionality poses challenges for accurate gene ranking.
• Existing gene ranking methods struggle with the complexity and scale of modern RNA-seq datasets.

Purpose of the Study:

• To propose and evaluate a novel, accurate, and sensitive gene ranking method for RNA-seq data using Discriminant Non-negative Matrix Factorization (DNMF).
• To demonstrate the superiority of DNMF over existing methods in identifying differentially expressed genes.

Main Methods:

• Implemented Discriminant Non-negative Matrix Factorization (DNMF) by incorporating Fisher's discriminant criteria with a reduced dimension of two.
• DNMF learns two factors representing metagenes (up-regulated/down-regulated patterns) and their expression values using sample labels.
• Genes are ranked based on the differential values of learned metagene weights.

Main Results:

• DNMF significantly outperformed widely used gene ranking methods in Area Under the Curve analysis on benchmark RNA-seq datasets.
• Gene Set Enrichment Analysis confirmed DNMF's superior performance.
• DNMF demonstrated substantial computational efficiency compared to other methods.

Conclusions:

• DNMF is an effective and robust method for differential gene expression analysis and gene ranking in RNA-seq data.
• The proposed DNMF approach offers improved accuracy, sensitivity, and computational speed for transcriptomic studies.