Enhanced drug classification using machine learning with multiplexed cardiac contractility assays.

Area of Science:

• Biomedical Engineering
• Cardiovascular Pharmacology
• Stem Cell Biology

Background:

• Drug-induced cardiotoxicity is a major cause of clinical trial failure.
• Current preclinical models inadequately predict human cardiac response.
• Human pluripotent stem cell-derived cardiomyocytes offer a promising alternative for cardiac safety assessment.

Purpose of the Study:

• To develop and validate an engineered cardiac tissue model for predicting drug cardiotoxicity.
• To create a robust dataset of electrophysiology and contractility responses to known compounds.
• To establish an ensemble algorithm for classifying the mechanistic action of unknown compounds.

Main Methods:

• Utilized three distinct engineered cardiac tissue assays.
• Exposed tissues to increasing concentrations of compounds from 5 mechanistic classes.
• Developed an ensemble algorithm integrating data from six individual models.

Main Results:

• Generated a comprehensive dataset on cardiac electrophysiology and contractility.
• The ensemble algorithm achieved 86.2% predictive accuracy in classifying compound mechanistic action.
• Outperformed predictions from single-assay models.

Conclusions:

• Engineered cardiac tissue provides a more predictive preclinical model for human cardiac response.
• The ensemble algorithm enhances the accuracy of drug safety screening.
• This approach aligns with regulatory advancements like the FDA Modernization Act 2.0 to improve drug development success rates.