Multi-labeled neural network model for automatically processing cardiomyocyte mechanical beating signals in drug assessment

Abstract

High-throughput cardiotoxicity assessment is important for large-scale preclinical screening in novel drug development. To improve the efficiency of drug development and avoid drug-induced cardiotoxicity, there is a huge demand to explore the automatic and intelligent drug assessment platforms for preclinical cardiotoxicity investigations. In this work, we proposed an automatic and intelligent strategy that combined automatic feature extraction and multi-labeled neural network (MLNN) to process cardiomyocytes mechanical beating signals detected by an interdigital electrode biosensor for the assessment of drug-induced cardiotoxicity. Taking advantages of artificial neural network, our work not only classified different drugs inducing different cardiotoxicities but also predicted drug concentrations representing severity of cardiotoxicity. This has not been achieved by conventional strategies like principal component analysis and visualized heatmap. MLNN analysis showed high accuracy (up to 96%) and large AUC (more than 98%) for classification of different drug-induced cardiotoxicities. There was a high correlation (over 0.90) between concentrations reported by MLNN and experimentally treated concentrations of various drugs, demonstrating great capacity of our intelligent strategy to predict the severity of drug-induced cardiotoxicity. This new intelligent bio-signal processing algorithm is a promising method for identification and classification of drug-induced cardiotoxicity in cardiological and pharmaceutical applications.

Keywords: Automatic feature extraction; Automatic segmentation; Cardiomyocytes; Interdigital electrode biosensor; Multi-labeled neural network.