Enhanced drug classification using machine learning with multiplexed cardiac contractility assays

Abstract

Cardiac screening of newly discovered drugs remains a longstanding challenge for the pharmaceutical industry. While therapeutic efficacy and cardiotoxicity are evaluated through preclinical biochemical and animal testing, 90 % of lead compounds fail to meet safety and efficacy benchmarks during human clinical trials. A preclinical model more representative of the human cardiac response is needed; heart tissue engineered from human pluripotent stem cell derived cardiomyocytes offers such a platform. In this study, three functionally distinct and independently validated engineered cardiac tissue assays are exposed to increasing concentrations of known compounds representing 5 classes of mechanistic action, creating a robust electrophysiology and contractility dataset. Combining results from six individual models, the resulting ensemble algorithm can classify the mechanistic action of unknown compounds with 86.2 % predictive accuracy. This outperforms single-assay models and offers a strategy to enhance future clinical trial success aligned with the recent FDA Modernization Act 2.0.

Keywords

• Cardiac tissue engineering
• Pharmaceutical compound testing
• Drug discovery
• Cardiac contractility
• Electrophysiology
• In vitro screening