Addressing the mean-correlation relationship in co-expression analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene co-expression analysis uses gene correlations to build gene networks from expression data.
• A mean-correlation relationship exists in RNA-seq data, where highly expressed genes show higher correlations.
• This bias is technical, not biological, and can obscure important low-expression gene interactions.

Purpose of the Study:

• To address the technical bias in gene co-expression network analysis caused by expression-dependent correlations.
• To develop a normalization method that corrects for the mean-correlation relationship in RNA-seq data.
• To improve the accuracy of gene network reconstruction by mitigating expression bias.

Main Methods:

• Introduction of spatial quantile normalization (SpQN), a novel method for normalizing local distributions within correlation matrices.
• Application of SpQN to RNA-seq data (both bulk and single-cell) to analyze gene co-expression patterns.
• Comparison of SpQN-corrected networks with uncorrected networks and protein-protein interaction data.

Main Results:

• Spatial quantile normalization effectively removes the mean-correlation relationship observed in RNA-seq data.
• SpQN corrects the expression bias, ensuring that correlations are less dependent on gene expression levels.
• The method enhances the accuracy of gene network reconstruction, particularly for lowly expressed genes.

Conclusions:

• SpQN is a valuable tool for correcting technical biases in gene co-expression network analysis.
• By removing the mean-correlation relationship, SpQN facilitates the identification of biologically relevant gene interactions.
• This normalization method improves the reliability of gene networks derived from RNA-seq data.