KG-Predict

Overview

KG-Predict is a computational framework designed for drug repurposing through knowledge graph embedding and inference. The system addresses the challenge of integrating and analyzing complex, heterogeneous biomedical data to identify new therapeutic applications for existing drugs.

The framework constructs GP-KG (Genotype-Phenotype Knowledge Graph), a comprehensive knowledge graph that integrates multiple types of entities and relations from various genotypic and phenotypic databases. GP-KG contains 1,246,726 associations between 61,146 biomedical entities, including drugs, diseases, genes, proteins, pathways, and phenotypes.

KG-Predict aggregates heterogeneous topological and semantic information from GP-KG to learn low-dimensional representations of entities and relations using advanced graph embedding techniques. These learned representations enable the inference of novel drug-disease interactions for drug repurposing applications.

Key Features

• Comprehensive Knowledge Graph: GP-KG with 1,246,726 associations and 61,146 entities
• Multi-modal Integration: Combines genotypic and phenotypic data from multiple databases
• Graph Embedding: Advanced methods for learning entity and relation representations
• High Performance: AUROC = 0.981, AUPR = 0.409, MRR = 0.261 in cross-validation
• Validated Applications: Successfully applied to Alzheimer’s disease drug discovery

Performance

In cross-validation experiments, KG-Predict demonstrated superior performance compared to other state-of-the-art graph embedding methods:

• AUROC (Area Under Receiver Operating Characteristic): 0.981
• AUPR (Area Under Precision-Recall): 0.409
• MRR (Mean Reciprocal Rank): 0.261

Case Study: Alzheimer’s Disease

KG-Predict was applied to identify novel repositioned candidate drugs for Alzheimer’s disease (AD), achieving:

• AUROC: 0.868
• AUPR: 0.364
• Successfully prioritized both FDA-approved and active clinical trial anti-AD drugs among top predictions

Data Sources

GP-KG integrates data from multiple genotypic and phenotypic databases, including information about:

• Drug-disease associations
• Drug-target interactions
• Gene-disease associations
• Protein-protein interactions
• Pathway information
• Phenotypic data

Technical Implementation

The framework is implemented in Python and includes:

• Configuration files for model parameters
• Data processing pipelines
• Graph embedding model implementations
• Training and testing scripts
• Case study analysis tools

Automated Evaluation

• View the automated evaluation: kg-predict automated evaluation