Machine learning of drug influence based on iPSC cardiomyocyte calcium transient signals

Martti Juhola1, Henry Joutsijoki1, Risto-Pekka Pölönen2, Katriina Aalto-Setälä1
1Tampere University, 2University of California Davis


Abstract

Calcium transient signal data of induced pluripotent stem cell-derived cardiomyocytes (iPSC-CM) originated from four individuals having a severe genetic arrhythmia. iPSC-CMs carrying mutations for catecholaminergic polymorphic ventricular tachycardia (CPVT) were used to study the effect of various drugs. Arrhythmias are induced in the patients with increasing of beating rate either with exercise or emotion. With cells, adrenaline was first used to increase beating and arrhythmias. After that various antiarrhythmic drugs were given to iPSC -CMs and the effect was analyzed. The response was analyzed with Ca2+-imaging technique. Peak detection was first executed for beats of calcium transient signals of cells exposed to various drugs by computing 14 different peak attributes from all valid peaks of every signal. The peak attribute data of approximately 150 signals were applied to classify calcium transient signals into different two or three classes depending on whether drug had an antiarrhythmic effect. Good antiarrhythmic response was obtained, if drug had modified “abnormal” peaks e.g. irregular, asymmetric peak shapes or peaks of various sizes to “normal” peaks i.e. mostly regular and symmetric shapes and almost similar sizes. Semi-response was obtained if some, but not all abnormalities were abolished and no-response was classified if the drug did not have any effect on the signal abnormalities. Classification was made with different machine learning methods giving various results. In the beginning, all signals had been annotated by an experienced researcher to three different classes along with potential drug influence (response, semi-response and no-response). When a machine learning method gave results clearly over 50% correctly, i.e., equal to those determined by expert researcher, this was seen as reasonable outcome. In the meeting we will present the data obtained in this project.