NECARE

NECARE predicts cancer-specific perturbations in protein–protein interaction (PPI) networks by integrating Relational Graph Convolutional Networks and knowledge-based features to identify cancer hub genes and prognostic network alterations.

Key Features:

• Relational Graph Convolutional Network (R-GCN): Uses R-GCNs to model complex, relation-specific interactions within cancer PPI networks.
• Knowledge-based features (OPA2Vec): Integrates OPA2Vec-derived features to incorporate structured biological knowledge into interaction prediction.
• Cancer-specific genomic features (TCGA): Utilizes mutation and expression profiles from The Cancer Genome Atlas (TCGA) to tailor predictions to cancer-specific molecular landscapes.
• Performance metrics: Demonstrates predictive performance with a Matthews correlation coefficient (MCC) of 0.84±0.03 and an F1 score of 91±2%.
• Experimental validation: Predictions were validated by coimmunoprecipitation experiments with 90% accuracy.

Scientific Applications:

• Mapping Cancer Interactome Atlas: Predicts PPIs to facilitate construction of a cancer interactome atlas that reveals network perturbations across cancer.
• Identification of cancer hub genes: Identifies 1,362 "cancer hub genes" that are significantly perturbed and enriched with mutations at protein–macromolecule binding interfaces.
• Prognostic relevance: Links hub genes to clinical relevance, noting that over 42% of treatment-related genes identified belong to these hub genes across 32 types of cancers.

Methodology:

NECARE trains Relational Graph Convolutional Networks on cancer-specific PPI data and integrates OPA2Vec-derived knowledge features with TCGA mutation and expression profiles; this approach addresses limitations of methods that relied on non-disease gene interaction data.