CardioGenAI: a machine learning-based framework for re-engineering drugs for reduced hERG liability.

Area of Science:

• Computational chemistry and drug discovery
• Cardiovascular safety pharmacology
• Machine learning in pharmaceutical research

Background:

• In vitro hERG ion channel inhibition correlates with in vivo QT interval prolongation, a risk for drug-induced arrhythmias like Torsade de Pointes.
• Early identification of hERG-active compounds is crucial to prevent the termination of promising drug candidates.
• Redesigning drugs to reduce hERG liability while preserving pharmacological activity is of significant interest.

Purpose of the Study:

• To present CardioGenAI, a machine learning framework for re-engineering drugs to reduce hERG activity.
• To predict activity against hERG, Na_V1.5, and Ca_V1.2 channels for comprehensive cardiovascular safety assessment.
• To demonstrate the framework's ability to optimize drug profiles while maintaining essential pharmacological and physicochemical properties.

Main Methods:

• Development of a machine learning-based framework (CardioGenAI) for drug re-engineering.
• Incorporation of state-of-the-art discriminative models for predicting ion channel activity (hERG, Na_V1.5, Ca_V1.2).
• Application of the framework to FDA-approved drugs, including pimozide, to generate refined candidates with improved safety profiles.

Main Results:

• CardioGenAI successfully generated 100 refined candidates from pimozide, including fluspirilene with 700-fold weaker hERG binding.
• The framework optimized hERG, Na_V1.5, and Ca_V1.2 profiles of multiple FDA-approved compounds.
• The redesigned compounds maintained their original physicochemical properties and pharmacological activity.

Conclusions:

• CardioGenAI offers a method to rescue drug development programs hindered by hERG-related safety concerns.
• The discriminative models within the framework can serve as standalone components for virtual screening.
• The open-source nature of CardioGenAI facilitates its integration into drug discovery workflows for molecular hypothesis generation.