Predicting adverse drug reactions through interpretable deep learning framework.

Area of Science:

• Computational chemistry
• Pharmacology
• Drug discovery

Background:

• Adverse drug reactions (ADRs) are unintended and harmful effects from normal drug use.
• Early prediction of ADRs is crucial for enhancing drug safety and reducing development costs.

Purpose of the Study:

• To develop a machine learning model for simultaneous ADR prediction and molecular substructure identification.
• To identify key molecular substructures linked to specific ADRs without prior definition.

Main Methods:

• Development of a deep learning framework for predicting ADRs.
• Simultaneous identification of molecular substructures associated with ADRs.
• Evaluation against ten state-of-the-art fingerprint models.

Main Results:

• The deep learning model, utilizing neural fingerprints, outperformed existing methods in ADR prediction.
• Identification of critical molecular substructures associated with specific ADRs through feature analysis.
• Statistical assessment confirmed the associations between identified substructures and ADRs.

Conclusions:

• The developed deep learning model offers a robust solution for identifying risky molecular components.
• This approach enhances drug safety evaluation by integrating substructure identification and statistical analysis.
• The model shows promise for improving the overall safety assessment in drug development.