HypergraphSynergy

HypergraphSynergy predicts synergistic effects of anti-cancer drug combinations by learning multi-way relations in a hypergraph of drugs and cancer cell lines to prioritize combinations for experimental validation.

Key Features:

• Hypergraph Representation: Models drugs and cancer cell lines as nodes and captures synergistic drug-drug-cell line triplets as hyperedges to represent complex multi-way relationships.
• Biochemical Feature Integration: Incorporates biochemical features of both drugs and cell lines as node attributes within the hypergraph.
• Hypergraph Neural Network (HGNN): Employs a specialized hypergraph neural network to learn embeddings from the hypergraph structure for synergy prediction.
• Auxiliary Task for Generalization: Includes an auxiliary task that reconstructs similarity networks of drugs and cell lines to refine embeddings and improve generalization.
• Prediction Tasks: Performs both classification and regression predictions of drug synergy on benchmark datasets.

Scientific Applications:

• Oncology research: Predicts synergistic effects of drug combinations across cancer cell lines to prioritize promising pairs for experimental validation and study of combination therapies.

Methodology:

Formulates drug synergy data as a hypergraph with drugs and cell lines as nodes and drug-drug-cell line hyperedges, incorporates biochemical node attributes, learns node and hyperedge embeddings using a hypergraph neural network, employs an auxiliary task to reconstruct drug and cell line similarity networks, and predicts synergy via classification and regression on benchmark datasets.