PheGenEx

PheGenEx applies the rcNet relationship-centric network approach to predict and prioritize phenotype–gene associations for human diseases, facilitating interpretation of candidate genes from GWAS, DNA copy number variation, and gene expression profiling.

Key Features:

• Network-based rcNet: Uses the rcNet relationship-centric network to model relationships between gene sets and disease phenotypes and to leverage known phenotype–gene associations.
• Learning framework: Employs a learning framework that maximizes coherence between predicted phenotype–gene set relations and established associations.
• Algorithmic design: Integrates ridge regression with label propagation in an efficient algorithm, implemented in two variants to optimize the learning objective.
• Validation and performance evaluation: Validated by leave-one-out cross-validation on OMIM and an independent test set of recently discovered disease–gene associations, achieving top rankings relative to baseline methods.
• Application to novel candidate genes: Applied to prioritize target diseases for novel candidate genes from GWAS, DNA copy number variation analysis, and gene expression profiling, including poorly annotated genes.

Scientific Applications:

• Candidate gene validation: Validates candidate disease genes identified by high-throughput genomic studies.
• Target disease prioritization: Prioritizes target diseases for candidate genes from GWAS, DNA copy number variation, and gene expression profiling.
• Poorly annotated gene interpretation: Infers phenotype associations for poorly annotated genes using the network-based rcNet approach with label propagation.
• Disease genetics and therapeutic target discovery: Supports investigation of the genetic basis of human diseases and identification of potential therapeutic targets.
• Personalized medicine and diagnosis: Contributes to personalized medicine by informing disease diagnosis and treatment strategy development.

Methodology:

Constructs a network of known phenotype–gene associations and applies the rcNet relationship-centric network with a learning framework that maximizes coherence between predicted and established relations; the algorithm integrates ridge regression with label propagation (two variants) and was evaluated by leave-one-out cross-validation on OMIM and an independent test set of recently discovered disease–gene associations.