KGML-xDTD: a knowledge graph-based machine learning framework for drug treatment prediction and mechanism description

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence in medicine

Background:

• Computational drug repurposing offers a cost-effective alternative to traditional drug discovery, particularly for rare diseases.
• A key challenge is the lack of understanding of the mechanisms of action (MOAs) for repurposed drugs.
• This knowledge gap hinders the clinical adoption of computational drug repurposing methods.

Purpose of the Study:

• To develop a novel framework, KGML-xDTD, for explainable prediction of drugs treating diseases.
• To provide biologically interpretable explanations for drug repurposing predictions.
• To enhance the confidence and accelerate the adoption of computational drug repurposing.

Main Methods:

• Developed a Knowledge Graph-based Machine Learning (KGML) framework named KGML-xDTD.
• Employed a two-module system for predicting treatment probabilities and generating explanations.
• Utilized Graph-based Reinforcement Learning (GRL) with knowledge- and publication-derived "demonstration paths" for mechanism explanation.

Main Results:

• KGML-xDTD achieved state-of-the-art performance in predicting drug repurposing.
• The framework successfully recapitulated human-curated drug MOA paths.
• Demonstrated the ability to provide testable, KG path-based explanations for predictions.

Conclusions:

• KGML-xDTD is the first framework to offer KG path explanations for drug repurposing predictions.
• The model addresses "black-box" concerns, increasing confidence in computational predictions.
• Facilitates faster drug discovery for emerging diseases by providing explainable insights.