Adverse Drug Reaction Discovery Using a Tumor-Biomarker Knowledge Graph

Area of Science:

• Biomedical Informatics
• Computational Biology
• Pharmacovigilance

Background:

• Adverse drug reactions (ADRs) pose significant risks to public health and drug development.
• Early detection of ADRs is critical for patient safety and effective therapeutic strategies.
• Biomedical literature mining offers a rich source for identifying potential ADRs.

Purpose of the Study:

• To develop and validate a machine learning approach for predicting unknown adverse drug reactions (ADRs) from biomedical literature.
• To construct an explainable Tumor-Biomarker Knowledge Graph (TBKG) for ADR discovery.
• To identify potential ADRs of antitumor drugs and provide mechanistic insights.

Main Methods:

• Construction of a Tumor-Biomarker Knowledge Graph (TBKG) integrating tumor, biomarker, drug, and ADR information from biomedical literature.
• Application of machine learning algorithms to the TBKG for predicting novel ADRs associated with antitumor drugs.
• Validation of predicted ADRs using real-world data, including clinical validation for Osimertinib.

Main Results:

• The developed model achieved 0.81 accuracy in three cross-validation tests.
• Predicted ADRs for Osimertinib included known reactions and novel, unreported rare ADRs.
• The model outperformed traditional co-occurrence methods in ADR discovery.
• The knowledge graph provided "tumor-biomarker-drug" paths, offering explanations for predicted ADRs.

Conclusions:

• The tumor-biomarker knowledge graph-based approach provides an explainable method for discovering potential ADRs.
• This method aids in understanding the mechanisms underlying ADRs and enhances patient safety.
• The approach is valuable for biomedical literature mining and advancing ADR research.