CoVar: A generalizable machine learning approach to identify the coordinated regulators driving variational gene...

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• Network inference models biological interactions (transcriptional, signaling, metabolic) to understand disease pathogenesis.
• Machine learning (ML)-based inference models show promise in uncovering complex patterns in genomic data.
• Existing statistical models often struggle to identify causative factors in complex diseases.

Purpose of the Study:

• To introduce CoVar, an ML framework for identifying central genes driving perturbed gene expression.
• To differentiate CoVar's approach from traditional differentially expressed genes (DEGs) analysis by focusing on network interaction changes.
• To reveal regulatory dynamics and coordinated processes underlying gene expression changes in biological states.

Main Methods:

• CoVar utilizes ML to identify variational genes with altered expression network interaction profiles.
• It further identifies core genes among the nearest neighbors of variational genes.
• The framework was validated using simulated data and yeast expression data with mitochondrial genome deletion.

Main Results:

• CoVar effectively captures intrinsic variationality and modularity in gene expression data.
• It successfully identifies key driver genes missed by existing differential analysis methods.
• The study demonstrates CoVar's ability to provide insights into regulatory dynamics.

Conclusions:

• CoVar offers a novel approach to understanding gene regulatory networks and disease mechanisms.
• Identifying variational and core genes provides a deeper understanding of coordinated regulatory processes.
• This ML-based framework advances the analysis of genomic data for complex disease research.