Inferring gene regulatory relationships with a high-dimensional robust approach

Area of Science:

• Genomics
• Bioinformatics
• Cancer Biology

Background:

• Gene expression (GE) levels are critical in biology and medicine, influenced by copy number alterations (CNAs).
• Understanding GE-CNA regulatory networks is vital for disease biology and translational medicine.
• Existing models struggle with long-tailed distributions and data contamination common in GE data.

Purpose of the Study:

• To develop a high-dimensional robust regression approach for inferring GE-CNA regulatory relationships.
• To address limitations of existing methods by accommodating cis- and trans-acting CNAs and robustly handling data contamination.
• To apply the novel method to analyze The Cancer Genome Atlas (TCGA) cutaneous melanoma data.

Main Methods:

• Implemented a high-dimensional regression model incorporating both cis- and trans-acting CNAs.
• Utilized a density power divergence loss function for robustness against long-tailed distributions and contamination.
• Employed penalization for regularized estimation and selection of relevant CNAs, realized via a coordinate descent algorithm.

Main Results:

• The proposed robust regression approach demonstrated competitive performance against nonrobust and LAD benchmarks in simulations.
• Analysis of TCGA cutaneous melanoma data revealed significant GE-CNA regulations within the regulation of apoptosis (RAP) pathway.
• The method successfully identified regulatory relationships, confirming its satisfactory performance on real-world cancer data.

Conclusions:

• The developed high-dimensional robust regression method effectively infers GE-CNA regulatory relationships.
• This approach offers improved accuracy and reliability for analyzing complex genomic data, particularly in the presence of noise and contamination.
• The findings provide valuable insights into cancer biology and potential therapeutic targets.