PerturbNet

PerturbNet models gene networks that mediate effects of single nucleotide polymorphisms (SNPs) on clinical phenotypes.

Key Features:

• Probabilistic Graphical Modeling: PerturbNet employs a probabilistic graphical model to represent the cascade from SNPs through gene networks to phenotypic outcomes.
• Comprehensive Network Learning: It learns entire gene network structures and identifies network modules that mediate SNP effects on phenotypes.
• Unified Optimization Algorithm: It solves a unified optimization problem with an efficient algorithm to enable genome-wide analysis, typically within a few hours.
• Enhanced Statistical Power: By modeling gene-network modulation of genetic effects, it increases statistical power to detect disease-linked SNPs in simulated and real datasets, including asthma studies.
• Multi-layer Integration: The framework integrates data across biological layers to enable simultaneous inference of genetic effects, gene-network structure, and phenotype-network relationships.

Scientific Applications:

• Systems Biology of Complex Diseases: Applied in systems biology and genomics to dissect molecular mechanisms underlying complex diseases.
• Network-mediated SNP Mapping: Used to identify gene networks and modules that mediate SNP effects on clinical traits.
• Improved Detection of Disease-associated SNPs: Leveraged to increase power to detect disease-linked SNPs by incorporating network structure.
• Pathophysiology Interpretation and Therapeutic Insight: Used to aid interpretation of pathophysiological mechanisms and to inform potential therapeutic strategies.
• Empirical Demonstration: Demonstrated on simulated data and real-world applications, including asthma studies.

Methodology:

Constructs a probabilistic model capturing the perturbation cascade from SNPs through gene networks to clinical phenotypes, integrates data across biological layers to simultaneously infer genetic effects, gene-network and phenotype-network structures, and uses an efficient optimization algorithm to solve a unified optimization problem for large-scale genomic datasets.