Artificial Intelligence and Machine Learning Methods to Evaluate Cardiotoxicity following the Adverse Outcome Pathway...

Area of Science:

• Toxicology
• Computational Chemistry
• Pharmacology

Background:

• Cardiovascular disease is a major global health concern.
• Chemicals like environmental contaminants, pesticides, and drugs can cause cardiotoxicity through various Adverse Outcome Pathways (AOPs).
• Synergistic effects between chemicals complicate hazard assessment.

Purpose of the Study:

• To employ Artificial Intelligence (AI) and Machine Learning (ML) for assessing chemical cardiotoxicity.
• To analyze chemical interactions with biological targets within AOP networks.
• To develop and evaluate advanced methods for encoding chemical information for AI/ML models.

Main Methods:

• Utilized ML and AI to model chemical interactions within cardiotoxicity AOPs, from molecular initiating events (MIEs) to key events (KEs).
• Explored various chemical encoding strategies, including molecular descriptors, fingerprints, graph-based methods, auto-encoders, and character embeddings.
• Developed a multimodal neural network architecture to integrate diverse chemical representations.

Main Results:

• Demonstrated the effectiveness of AI/ML in predicting chemical cardiotoxicity.
• Showcased advanced chemical encoding methods for improved model performance.
• Highlighted the advantages of multimodal architectures, especially with increasing data volumes.

Conclusions:

• AI and ML offer powerful in silico tools for cardiotoxicity assessment, complementing traditional methods.
• Advanced chemical representations and multimodal networks enhance the accuracy and efficiency of hazard evaluation.
• This approach supports Integrated Approaches to Testing and Assessment (IATA), reducing reliance on in vivo studies.