Predicting Drugs Suspected of Causing Adverse Drug Reactions Using Graph Features and Attention Mechanisms

Area of Science:

• Pharmacovigilance
• Computational Chemistry
• Drug Discovery

Background:

• Adverse drug reactions (ADRs) are unintended harmful effects of medications, causing significant hospital admissions and healthcare costs.
• Increasing awareness of ADRs is crucial for prevention, with drug assessment in adverse events being a key strategy.
• ADRs are a major public health concern, necessitating advanced methods for identification and prediction.

Purpose of the Study:

• To develop and validate a novel model for identifying suspected drugs associated with adverse events.
• To predict potential adverse drug reactions (ADRs) for various medications.
• To explore the model's utility in broader drug discovery and classification tasks.

Main Methods:

• A suspect drug assisted judgment model (SDAJM) was designed using graph isomorphism network (GIN) and attention mechanisms.
• The model extracts features from patient demographics, drug information, and adverse drug reaction data.
• Feature extraction was performed to identify patterns indicative of drug-induced adverse events.

Main Results:

• The SDAJM demonstrated strong performance in predicting suspected drugs in adverse reaction events compared to other models.
• Case analyses on cardiovascular and antithyroid drugs confirmed the model's capability in predicting drug-induced ADRs.
• Validation on benchmark datasets (Tox21, SIDER) confirmed the model's applicability to classification tasks in drug discovery.

Conclusions:

• The SDAJM, utilizing deep learning, effectively identifies drugs causing adverse drug events (ADEs).
• The model contributes to predicting drug ADRs and supports other drug discovery endeavors.
• This research offers novel approaches for advancing the field of ADR research and pharmacovigilance.