top of page
Gradient
Search
Writer's picturekübra:)

Machine Learning Techniques to Classify Healthy and Diseased Cardiomyocytes by Contractility Profile

Cardiomyocytes derived from human induced pluripotent stem (iPS) cells enable the study of cardiac physiology and the developmental testing of new therapeutic drugs in a human setting. In parallel, machine learning methods are being applied to biomedical science in unprecedented ways. Machine learning has been used to distinguish healthy from diseased cardiomyocytes using calcium (Ca2+) transient signals. Most Ca2+ transient signals are obtained via terminal assays that do not permit longitudinal studies, although some recently developed options can circumvent these concerns. Here, they describe the use of machine learning to identify healthy and diseased cardiomyocytes according to their contractility profiles, which are derived from brightfield videos. This noncontact, label-free approach allows for the continued cultivation of cells after they have been evaluated for use in other assays and can be readily extended to organs-on-chip. To demonstrate utility, they assessed contractility profiles of cardiomyocytes obtained from patients with Timothy Syndrome (TS), a long QT disease which can lead to fatal arrhythmias, and from healthy individuals. The videos were processed and classified using machine learning methods and their performance was evaluated according to several parameters. The trained algorithms were able to distinguish the TS cardiomyocytes from healthy controls and classify two different healthy controls. The proposed computational machine learning evaluation of human iPS cell-derived cardiomyocytes’ contractility profiles has the potential to identify other genetic proarrhythmic events, screen therapeutic agents for inducing or suppressing long QT events, and predict drug–target interactions. The same approach could be readily extended to the evaluation of engineered cardiac tissues within single-tissue and multi-tissue organs-on-chip.


Graphical Abstract


Cardiomyocytes derived from human induced pluripotent stem (iPS) cells are finding utility in the discovery of new therapeutic agents and in the modeling of human diseases in vitro. Because human iPS cells, when differentiated, retain the original genotype from the cell donor, they are increasingly used in studies that go beyond measuring mere cardiac functionality, into the realm of modeling human cardiovascular diseases, such as long QT syndrome, myocarditis, acute ischemia, and further into high-throughput cardiotoxicity screening.


Experimental overview. (A) Human induced pluripotent stem (iPS) cells were differentiated into cardiomyocytes from three different cell lines. After differentiation, their contractility profiles were evaluated using brightfield videos. (B) Contractility trace obtained from brightfield videos. (C) The performance of the algorithms was evaluated using parameters calculated from their confusion matrixes and receiver operating characteristic (ROC) curves.


An example of successful recapitulation of human disease in vitro by human iPS cell-derived cardiomyocytes is Timothy Syndrome (TS), a disease characterized by prolonged QT intervals. Patients with TS carry a spontaneous autosomal dominant gain-of-function mutation in the CACNA1C gene encoding Cav1.2 channels. Two known effects of this mutation are the slower inactivation of the ion channels, resulting in prolongation of the QT interval, and cardiac arrhythmia that can lead to sudden cardiac death. TS patients commonly exhibit bradycardia, an outcome that has been replicated in vitro using iPS cell-derived cardiomyocytes from affected patients.


Contractility parameters. The eight parameters were calculated from the contractility traces of cardiomyocytes differentiated from three human iPS cell lines. Data is presented as mean + standard deviation. Differences between experimental groups were analyzed by Kruskal–Wallis test, followed by Dunn’s multiple comparisons test. Significant differences are defined by P < 0.05 (*), P < 0.01 (**), P < 0.001 (***), and P < 0.0001 (****).


Machine learning, the process of training an algorithm to make predictions or decisions based on experimental data, has been used to process multidimensional datasets in an objective and automated fashion, providing the opportunity to store and analyze large datasets quickly, rather than having to manually preselect a limited number of parameters and thereby overlooking potentially valuable information. Supervised machine learning is a subtype of machine learning in which a set of data with known classifications is used to train an algorithm by building a statistical model that fits the data. This trained model can then be applied to unknown data to predict their classification and performance.



ROC curves of the algorithms with the highest AUC. (A) Classification of WTC-11 and TS. (B) Classification of BS2 and TS. (C) Classification of BS2 and WTC-11.


One of the current challenges in treating cardiac disease and in the development of new therapeutic agents is the need for their accurate and fast preclinical detection and screening. By integrating machine learning techniques with current models, preclinical drug screening and disease modeling can be accelerated in an automated, easy-to-use fashion. Machine learning algorithms can accelerate the classification of diseased cells, identify side effects of new cardioactive drugs under development, or evaluate the arrhythmic risk of patient-derived cells or cells exposed to new therapeutic agents.


Machine learning has been only rarely used for data obtained from human iPS cell-derived cardiomyocytes. Some groups have used machine learning techniques to predict the outcome of iPS cell differentiation protocols, while others focused on quality control of their cardiomyocyte cultures. Machine learning has also been used in the development of high-throughput and sensitive drug screening platforms and as an action potential classifier. Machine learning algorithms have been trained to identify peaks of calcium (Ca2+) transients in arrhythmogenic cardiomyocytes and the action potential of healthy cells exposed to antiarrhythmic drugs. One study introduced a method for automated analysis of the arrhythmic field potentials of cells exposed to cardioactive drugs, while another study reported the use of a platform paired with machine learning algorithms to detect changes in cardiac functionality after drug exposure.



Recently, healthy and diseased cardiomyocytes were separated by machine learning algorithms based on analysis of calcium transient signals. Calcium signaling plays an important role in cardiac functionality, both under healthy and pathological conditions. However, calcium transients are frequently obtained via terminal assays, preventing the use of the evaluated cells in future experiments and ongoing analysis. Some recently developed options can circumvent these concerns, but they are still not used routinely. Data obtained using a noncontact, online, label-free approach allows for the classification of cells without precluding their use in longitudinal studies (where the same cells are analyzed over time), in other assays, or in the screening of therapeutic agents. Machine learning algorithms can be further leveraged with new analysis tools in lieu of calcium signals from single cells. They previously developed a MATLAB script to analyze brightfield videos of beating cardiomyocytes and generate a contractility trace that can be used to calculate contractility parameters.


This approach enabled us to assess contractility profiles without the need to label or dissociate cells, allowing cell labeling for further analysis. They hypothesize that the contractility profiles obtained from brightfield videos can be used to reliably classify healthy and diseased cardiomyocytes. To test this hypothesis, they differentiated cardiomyocytes from three cell lines (two healthy and one from a TS patient). Their contractility traces were extracted from brightfield videos and analyzed using a custom MATLAB script. The calculated contractility parameters served as a data input to several machine learning algorithms that were trained to distinguish the contractile behaviors of diseased and healthy cells. They propose that these algorithms for automated analysis of contractility profiles can be used to detect pathologic phenotypes and evaluate therapeutic agents.

  1. Machine Learning Techniques to Classify Healthy and Diseased Cardiomyocytes by Contractility Profile Diogo Teles, Youngbin Kim, Kacey Ronaldson-Bouchard, and Gordana Vunjak-Novakovic ACS Biomaterials Science & Engineering 2021 7 (7), 3043-3052 DOI: 10.1021/acsbiomaterials.1c00418

Comments


bottom of page