CardioDPi-Predictor：A Web-server for Prediction of Cardiotoxic Compounds using Deep Neural Network Model
Abstract
Toxic compounds in the environment pose a significant threat to human health and are major contributors to many chronic diseases, especially cardiovascular diseases. We have developed an easy-to-use CardioDPi system, which is an integrated single-model of hERG, Nav1.5, and Cav1.2 channels employing a deep learning approach based on fully connected neural networks combined with molecular fingerprints for predicting the cardiac toxicity of compounds affecting different ion channels. It has a broader applicability and each provides enhanced predictive capability. The model evaluation metrics are displayed in the following table.
| Metrics | hERG | Nav1.5 | Cav1.2 |
| AUC | 0.89 | 0.89 | 0.94 |
| ACC | 0.82 | 0.86 | 0.86 |
| SE | 0.83 | 0.90 | 0.87 |
| SP | 0.81 | 0.78 | 0.85 |
| F1-score | 0.82 | 0.89 | 0.89 |