Predicting Adverse Drug Effects from Literature- and Database-Mined Assertions.

Area of Science:

• Pharmacology and Cheminformatics
• Biomedical Data Mining
• Drug Safety and Development

Background:

• Adverse drug effects (ADEs) cause significant patient harm and drug withdrawals.
• Early prediction of drug-target-effect relationships is crucial for drug development.
• Integrating diverse pharmacological data can facilitate relationship inference.

Purpose of the Study:

• To identify known and unknown relationships between chemicals (C), protein targets (T), and ADEs (E) using literature evidence.
• To develop a computational workflow for predicting potential ADEs.
• To validate the predictive capability of the inference strategy.

Main Methods:

• Employed cheminformatics and data mining to integrate public clinical pharmacology data and literature assertions.
• Developed a C-T-E relationship knowledge base and formed C-T-E triangles.
• Inferred missing C-E edges (relationships) as potential ADEs.

Main Results:

• Inferred novel, unreported associations between drugs, targets, and ADEs.
• Prioritized inferences as testable hypotheses, including testosterone → myocardial infarction.
• Timestamping confirmed the predictive accuracy of the inference strategy.

Conclusions:

• The workflow, using free databases and an inference scheme, identified novel C-E relationships validated by case reports.
• This computational method offers potential for early detection of drug candidate ADEs.
• Refinement of prioritization schemes can enhance the workflow's effectiveness for targeted experimental investigation.