Cheminformatics-aided pharmacovigilance: application to Stevens-Johnson Syndrome

Area of Science:

• Pharmacovigilance
• Computational Chemistry
• Drug Safety

Background:

• Adverse Drug Reactions (ADRs) pose significant patient safety risks.
• Early identification of ADRs, such as Stevens-Johnson Syndrome (SJS), is crucial for patient protection.
• Quantitative Structure-Activity Relationship (QSAR) models offer a predictive approach to drug safety.

Purpose of the Study:

• To investigate if global spontaneous reporting patterns can identify chemical substructures linked to SJS.
• To develop and validate QSAR models for predicting SJS association in drugs.
• To utilize QSAR models for early warning of potential drug-induced SJS.

Main Methods:

• Utilized 364 drugs from VigiBase with known SJS reporting correlations.
• Computed chemical descriptors from drug molecular structures.
• Developed and validated QSAR models using Random Forest and Support Vector Machines, with external cross-validation.

Main Results:

• Developed QSAR models with high accuracy (AUC 75%-81%) for predicting SJS association.
• Successfully predicted SJS active and inactive drugs through virtual screening of DrugBank.
• Validated predictions against literature and knowledge bases, identifying novel SJS structural alerts.

Conclusions:

• QSAR models accurately identify drugs associated with SJS.
• These models provide an effective computational tool for flagging potentially harmful drugs.
• The approach supports targeted surveillance and pharmacoepidemiologic investigations.