Robust network-based analysis of the associations between (epi)genetic measurements

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Modeling gene expression (GE) and copy number variation (CNV) associations is crucial but challenging due to high dimensionality, CNV regulation complexity, cis/trans effects, data contamination, and CNV correlations.
• Existing methods often fail to address multiple challenges simultaneously, limiting their effectiveness in understanding GE-CNV relationships.

Purpose of the Study:

• To develop a novel, robust method for modeling gene expression and copy number variation associations.
• To address key challenges including high dimensionality, non-linear cis-acting effects, data contamination, and correlations among CNVs.

Main Methods:

• A partially linear model with a non-linear cis-acting CNV effect is proposed for each gene expression.
• A robust loss function is employed to handle long-tailed distributions and data contamination.
• Penalization is used for high dimensionality and relevant CNV identification, while a network structure accounts for CNV correlations.

Main Results:

• The proposed method effectively accommodates multiple challenging characteristics of GE-CNV modeling, overcoming limitations of existing approaches.
• Simulation studies demonstrate the superiority of the new method compared to alternatives.
• Application to The Cancer Genome Atlas (TCGA) programmed cell death (PCD) pathway data yielded improved prediction, stability, and biologically plausible findings.

Conclusions:

• The developed method provides a comprehensive and effective framework for modeling GE-CNV associations.
• It offers improved analytical power and biological interpretability for genomic data, particularly in cancer research.
• This approach advances the understanding of how copy number variations influence gene expression patterns.