A powerful framework for differential co-expression analysis of general risk factors

Area of Science:

• Bioinformatics
• Systems Biology
• Genomics

Background:

• Differential co-expression analysis (DCA) identifies genes with coordinated expression patterns influenced by risk factors.
• Current DCA methods are limited to categorical risk factors and susceptible to bias from batch and variance effects.

Purpose of the Study:

• To introduce Kernel-based Differential Co-expression Analysis (KDCA), a novel framework for detecting differential co-expression.
• To address limitations of existing DCA methods by accommodating general risk factors (continuous, discrete, categorical) and mitigating bias.

Main Methods:

• KDCA utilizes correlation patterns within gene pathways to detect differential co-expression.
• The framework was evaluated using simulated pathway data with diverse architectures.
• Performance was compared against the standard eigengene approach.

Main Results:

• KDCA effectively controls the type I error rate by accounting for common sources of bias.
• The method demonstrated substantially increased statistical power compared to the eigengene approach.
• Application to The Cancer Genome Atlas thyroid data identified differentially co-expressed pathways by age and BRAF mutation status, missed by the eigengene method.

Conclusions:

• KDCA is a powerful and versatile framework for differential co-expression analysis.
• The method expands the applicability of DCA in gene expression studies, particularly for complex risk factors and bias mitigation.