REDDA

REDDA predicts drug-disease associations using a relations-enhanced heterogeneous graph neural network that models drugs, diseases, genes, and proteins to support computational drug repositioning and discovery.

Key Features:

• Heterogeneous Graph Neural Network: Models complex biological interactions among drugs, diseases, genes, and proteins using a heterogeneous graph structure.
• Attention Mechanisms: Implements attention mechanisms comprising a general heterogeneous graph convolutional network-based node embedding block, a topological subnet embedding block, a graph attention block, and a layer attention block.
• Sequential Representation Learning: Sequentially learns and refines drug and disease representations through stacked attention and embedding modules.
• Integration of Biological Relations: Integrates multiple biological relations into a cohesive model to capture diverse interaction types relevant to association prediction.
• Performance Metrics: Outperformed eight advanced methods on benchmark datasets, achieving relative improvements of 0.76% AUC and 13.92% AUPR on one dataset and 2.48% AUC and 4.93% AUPR on another.
• Case Study Validation: Produced valid predictions in case studies that illustrate practical utility for drug-disease association discovery.

Scientific Applications:

• Computational Drug Repositioning: Predicts potential new indications for existing drugs and prioritizes candidate drug-disease associations for repurposing.
• In Silico Drug Discovery: Identifies novel therapies and supports exploration of drug-disease associations in pharmaceutical research, with validated predictions in case studies.

Methodology:

Integrates multiple biological relations (drugs, diseases, genes, proteins) within a heterogeneous graph neural network and applies GCN-based node embedding, topological subnet embedding, graph attention, and layer attention to learn and refine representations for association prediction.