Science Pool

Advances in Cardiotoxicity Prediction using Transcriptomics and Machine Learning

Cardiotoxicity is one of the leading causes of drug attrition. To address this, there is a need for improved predictive screens which can be applied at an early stage in drug development to ensure only safe compounds progress to the clinic.

Drug-induced cardiotoxicity can manifest itself via various different mechanisms which makes detection challenging. Either functional or structural changes can occur, and these effects can be direct or indirect. Functional cardiotoxicity results from acute alteration in the heart function often as a consequence of electrophysiological effects. Structural cardiotoxicity is associated with alterations in cell and tissue morphology and can manifest itself as cell death, inflammatory changes and fibrosis. These morphological changes can take time to present themselves clinically and, therefore, sensitive techniques which can detect early molecular changes are valuable from a predictive perspective.

Combining transcriptomics with artificial intelligence is showing great potential in providing this transformative improvement in predictive safety assessment. In fact, the value of this approach in drug-induced liver injury (DILI) has been demonstrated in a recent poster presented by Cyprotex. At SOT in Salt Lake City from 10-14 March 2024, Cyprotex showcased how this technique can also be applied to cardiotoxicity. The research evaluated 42 compounds (33 cardiotoxicants and 9 non-cardiotoxicants) used in a variety of therapeutic indications. The compounds were assessed at 2 time points and 8 concentrations in a beating cardiac organ model using human iPSC-derived cardiomyocytes. At the end of the incubation period, three different analytical methods were compared; high content screening (HCS) (cell count, cellular ATP, mitochondrial mass, mitochondrial membrane potential, cellular calcium levels, DNA structure and nuclear size), calcium transience (wave amplitude, frequency, full peak width and full decay time) and transcriptomics (high-throughput RNA sequencing).

The HCS and calcium transience assays were valuable in detecting structural and functional cardiotoxicants, respectively. However, the synergism of combining these assays with transcriptomics was demonstrated by an overall improved cardiotoxicity risk prediction. Additionally, transcriptomics analysis provided detailed mechanistic information and identified specific pathway responses. Combined, the three approaches (HCS, calcium transience and transcriptomics analysis) gave excellent cardiotoxicity prediction metrics of 100% specificity, 82% sensitivity and 86% accuracy at 10x Cmax and 89% specificity, 91% sensitivity and 90% accuracy at 25x Cmax. The transcriptomics analysis improved the overall sensitivity by identifying various molecular mechanisms of structural toxicity such as alterations in cardiac pathways, genotoxicity, ER stress and mitochondrial toxicity.

It is not just cardiomyocytes which can be affected by drug induced toxicity, non-cardiomyocytes such as fibroblasts and endothelial cells may also be impacted. Organotypic models developed using different cell types, therefore, are likely to be more representative both structurally and functionally. In the future, Cyprotex is extending its research to evaluate a number of different cell-based models and different organ-specific toxicities using the transcriptomics approach. Our cardiac safety database is also growing and we now have identified approximately 140 compounds for testing in our models. The use of transcriptomics and artificial intelligence is accelerating the development of new cell-based models in the field of drug-induced toxicity leading to a new paradigm in in vitro safety testing.

Interested in learning more?

CONTACT US to discuss your project

READ OUR PUBLICATION