Abstract:In the proposed intelligent biosensing drug‐induced cardiotoxicity assessment platform, the cardiomyocytes are treated with non‐cardiotoxic or cardiotoxic drugs to obtain mechanical beating signals. The beating signals are converted into image representations by phase space reconstruction and analyzed by convolutional neural network to identify the cardiotoxicity of drugs. Drug‐induced cardiotoxicity often leads to patient deaths and drug recalls. Interdigitated electrode (IDE)‐based cellular impedance detection instrument has been integrated with intelligent algorithms to screen cardiotoxicity based on cardiomyocytes. These intelligent biosensing systems generally employ traditional machine learning methods to assess drug‐induced cardiotoxicity by analyzing cardiomyocytes mechanical beating signals. However, modern deep learning methods with robustness and flexibility have not been integrated in IDE‐based platform to screen cardiotoxicity. Here, for the first time, we implemented deep convolutional neural network (CNN) to analyze cardiomyocytes mechanical beating signals for cardiotoxicity assessment. This method can eliminate the feature engineering procedures, such as manual design and extraction of signal features required by traditional machine learning methods. To facilitate two‐dimensional (2‐D) CNN analysis, we utilized phase space reconstruction to convert one‐dimensional beating signals into 2‐D image representations as well as take into account nonlinear dynamic information. The phase space reconstruction‐assisted convolutional neural network (PSRCNN) is capable of accurately categorizing drug‐induced cardiotoxicities and predicting cardiotoxicity levels. It obtains accuracies ranging from 0.82 to 0.99 when recognizing the cardiotoxicity of newly developed drugs, which are represented by drugs that are not used during model training. Furthermore, we explore in depth to compare the performance of CNN with other traditional machine learning methods. The PSRCNN method can achieve higher accuracies when compared to other methods. The PSRCNN‐based biosensing platform is highly potential in improving the efficiency and accuracy of high‐throughput screening of newly developed drugs for cardiotoxicity during drug discovery.

Multi-labeled Neural Network Model for Automatically Processing Cardiomyocyte Mechanical Beating Signals in Drug Assessment.

A biosensing system employing nonlinear dynamic analysis-assisted neural network for drug-induced cardiotoxicity assessment

An intelligent biosensing platform using phase space reconstruction‐assisted convolutional neural network for drug‐induced cardiotoxicity assessment

Early Detection of Coronary Microvascular Dysfunction Using Machine Learning Algorithm Based on Vectorcardiography and Cardiodynamicsgram Features

Electromechanical Integrated Recording of Single Cardiomyocyte in Situ by Multimodal Microelectrode Biosensing System.

A Biosensing System Using a Multiparameter Nonlinear Dynamic Analysis of Cardiomyocyte Beating for Drug-Induced Arrhythmia Recognition

Enhanced drug classification using machine learning with multiplexed cardiac contractility assays

Automated Heartbeat Classification Exploiting Convolutional Neural Network with Channel-Wise Attention

CardiOT: Towards Interpretable Drug Cardiotoxicity Prediction Using Optimal Transport and Kolmogorov-Arnold Networks

Cardiomyocyte electrical-mechanical synchronized model for high-content, dose-quantitative and time-dependent drug assessment

Deep Neural Network-Aided Histopathological Analysis of Myocardial Injury

A Multi-Label Learning Model for Predicting Drug-Induced Pathology in Multi-Organ Based on Toxicogenomics Data.

ToxMPNN: A Deep Learning Model for Small Molecule Toxicity Prediction.

Universal and Sensitive Drug Assessment Biosensing Platform Using Optimal Mechanical Beating Detection of Single Cardiomyocyte

Two-Dimensional Deep Learning Frameworks for Drug-Induced Cardiotoxicity Detection

A stacking ensemble machine learning model for evaluating cardiac toxicity of drugs based on in silico biomarkers

Multifunctional cardiac microphysiological system based on transparent ITO electrodes for simultaneous optical measurement and electrical signal monitoring

A deep learning based multi-model approach for predicting drug-like chemical compound's toxicity

Semi-Supervised Learning to Boost Cardiotoxicity Prediction by Mining a Large Unlabeled Small Molecule Dataset

A Deep Learning Based Multi-Model Approach for Predicting Drug-Like Chemical Compound’s Toxicity

A cardiomyocyte-based biosensor for antiarrhythmic drug evaluation by simultaneously monitoring cell growth and beating.