A powerful nonparametric method for detecting differentially co-expressed genes: distance correlation screening and...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Differential co-expression analysis complements differential expression analysis by revealing molecular mechanism changes in different phenotypes.
• Current methods often rely on Pearson correlation coefficients, which are limited in detecting nonlinear gene co-expression patterns common in gene regulatory networks.

Purpose of the Study:

• To propose a novel nonparametric procedure for identifying differentially co-expressed gene pairs across different phenotypes using large-scale data.
• To address the limitations of existing methods in detecting nonlinear co-expression changes.

Main Methods:

• A two-step computational pipeline involving a screening step and a testing step.
• Screening step utilizes distance correlation to filter independent gene pairs, reducing search space.
• Testing step employs a distribution-free edge-count test to compare gene co-expression patterns, focusing on nonlinear relations.

Main Results:

• The proposed method effectively identifies differentially co-expressed gene pairs, particularly those with nonlinear relationships.
• Analysis of Cancer Genome Atlas and METABRIC breast cancer data demonstrated the approach's utility.
• The distance correlation screening significantly enhances computational efficiency for large datasets.

Conclusions:

• The new nonparametric method offers superior power in detecting nonlinear differential co-expressions compared to existing techniques.
• The computational efficiency gained through distance correlation screening facilitates broader application in large-scale genomic studies.
• This approach provides valuable insights into the molecular mechanisms underlying different phenotypes.