KG4SL

KG4SL predicts synthetic lethality (SL) interactions in human cancers by integrating knowledge graphs with a graph neural network to capture complex biological relationships among genes, compounds, diseases, and biological processes.

Key Features:

• Graph Neural Network-Based Model: Employs a GNN framework to integrate topological and relational information from a knowledge graph for SL prediction.
• Knowledge Graph Integration: Constructs a knowledge graph comprising 11 types of entities and 24 kinds of relationships, enabling KG message-passing to represent interdependencies among genes, compounds, diseases, and biological processes.
• Avoidance of Manual Feature Engineering: Leverages KG message-passing within the GNN to learn predictive representations without domain-specific manual feature engineering.

Scientific Applications:

• Drug Target Discovery: Predicts SL interactions to identify candidate gene pairs exploitable as cancer drug targets.
• Reduction of Experimental Costs: Provides computational SL predictions to prioritize hypotheses and reduce the scale of wet-lab screening.

Methodology:

Implements a novel GNN-based model with knowledge graph message-passing to predict SL interactions; comparative experiments were performed against an unsupervised TransE model, a vanilla graph convolutional network, and their combination, and evaluated using AUC, AUPRC, and F1.