A Fusion Deep Learning Model for Predicting Adverse Drug Reactions Based on Multiple Drug Characteristics.

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Adverse drug reactions (ADRs) pose significant risks to patient safety and increase healthcare costs.
• Previous ADR prediction models often used limited data dimensions and focused on single ADRs per drug.
• Integrating diverse drug characteristics offers a more comprehensive approach to ADR prediction.

Purpose of the Study:

• To develop an AI-driven predictor for early identification of ADRs during drug discovery.
• To enhance ADR prediction accuracy by fusing multiple drug characteristics using a deep learning model.
• To address the limitation of predicting only single ADRs by enabling multi-label prediction.

Main Methods:

• Developed a deep learning model integrating four modules: 1D/2D drug molecular structure, drug-protein interactions, and drug similarity.
• Employed a fusion model to combine these characteristics for precise ADR probability prediction.
• Utilized benchmark and LIU's datasets for model evaluation and comparison with state-of-the-art methods.

Main Results:

• Achieved ROC-AUC of 0.7002, AUPR of 0.6619, and F1 score of 0.6330 on the benchmark dataset.
• Significantly improved AUPR compared to conventional multi-label classifiers (64.02% to 66.19%).
• Outperformed state-of-the-art methods on LIU's dataset, with AUPR increasing from 34.65% to 68.82%.

Conclusions:

• The developed AI model accurately predicts ADR probabilities by integrating comprehensive drug information.
• This approach offers significant value for enhancing drug safety monitoring in new drug development and clinical use.
• The model demonstrates superior accuracy and robustness in identifying potential adverse drug reactions.