MultiCTox: Empowering Accurate Cardiotoxicity Prediction through Adaptive Multimodal Learning

Area of Science:

• Pharmacology and Cheminformatics
• Computational Drug Discovery

Background:

• Cardiotoxicity, the adverse effect of drugs on cardiac function, poses a significant challenge in drug development.
• Current prediction methods often rely on single data types, limiting their accuracy and comprehensive analysis.
• Accurate cardiotoxicity prediction is essential for evaluating drug efficacy and ensuring patient safety.

Purpose of the Study:

• To develop an advanced multimodal method for enhanced prediction of drug-induced cardiotoxicity.
• To integrate diverse molecular data (SMILES, structure, fingerprint) for a more robust prediction model.

Main Methods:

• A novel fusion layer was designed to unify representations from molecular SMILES, structure, and fingerprint data.
• The model was trained to maximize intramodal similarity and minimize intermolecular similarity for consistent cross-modal representations.
• The method was evaluated by assessing inhibitory effects on key cardiac ion channels: hERG, Nav1.5, and Cav1.2.

Main Results:

• The proposed multimodal model demonstrated significantly superior performance compared to existing state-of-the-art cardiotoxicity prediction methods.
• Integration of multiple molecular data modalities led to enhanced prediction accuracy.
• The model effectively predicted drug effects on voltage-gated potassium (hERG), sodium (Nav1.5), and calcium (Cav1.2) channels.

Conclusions:

• The developed multimodal approach offers a substantial improvement in cardiotoxicity prediction accuracy.
• This method holds significant potential for advancing the safety evaluation of cardiac drugs.
• Implementing this model can help mitigate cardiotoxicity-related risks in drug development.