A scaling-free minimum enclosing ball method to detect differentially expressed genes for RNA-seq data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-sequencing (RNA-seq) data analysis requires normalization to account for technical effects and varying sequencing depths.
• Existing normalization methods may struggle with complex data exhibiting multiple scaling factors, potentially leading to suboptimal gene expression analysis.
• Machine learning offers new approaches, framing differential gene expression as a one-class classification problem.

Purpose of the Study:

• To develop a novel, normalization-free method for identifying differentially expressed genes (DEGs) in RNA-seq data.
• To address limitations of current methods that rely on single scaling factors for normalization.
• To provide a robust approach for cross-species gene expression comparisons.

Main Methods:

• Proposed a scaling-free minimum enclosing ball (SFMEB) method.
• Treated differentially expressed genes (DEGs) as outliers within a feature space.
• Constructed a minimum enclosing ball containing non-DE genes; genes outside the ball are identified as DEGs.

Main Results:

• The SFMEB method effectively identifies DEGs without requiring data normalization.
• Demonstrated superior performance compared to existing methods, particularly for heterogeneous data and biological replicates.
• Successfully applied to both same-species and cross-species gene expression analyses.

Conclusions:

• The SFMEB method is a robust and effective tool for identifying differentially expressed genes.
• Its ability to bypass normalization makes it advantageous for complex RNA-seq datasets.
• An R package for the SFMEB method is publicly available for broader research application.