Capsule Networks Showed Excellent Performance in the Classification of hERG Blockers/Nonblockers

Abstract

Capsule networks (CapsNets), a new class of deep neural network architectures proposed recently by Hinton et al., have shown a great performance in many fields, particularly in image recognition and natural language processing. However, CapsNets have not yet been applied to drug discovery-related studies. As the first attempt, we in this investigation adopted CapsNets to develop classification models of hERG blockers/ nonblockers; drugs with hERG blockade activity are thought to have a potential risk of cardiotoxicity. Two capsule network architectures were established: convolution-capsule network (Conv-CapsNet) and restricted Boltzmann machine-capsule networks (RBM-CapsNet), in which convolution and a restricted Boltzmann machine (RBM) were used as feature extractors, respectively. Two prediction models of hERG blockers/nonblockers were then developed by Conv-CapsNet and RBM-CapsNet with the Doddareddy's training set composed of 2,389 compounds. The established models showed excellent performance in an independent test set comprising 255 compounds, with prediction accuracies of 91.8 and 92.2% for Conv-CapsNet and RBM-CapsNet models, respectively. Various comparisons were also made between our models and those developed by other machine learning methods including deep belief network (DBN), convolutional neural network (CNN), multilayer perceptron (MLP), support vector machine (SVM), k-nearest neighbors (kNN), logistic regression (LR), and LightGBM, and with different training sets. All the results showed that the models by Conv-CapsNet and RBM-CapsNet are among the best classification models. Overall, the excellent performance of capsule networks achieved in this investigation highlights their potential in drug discoveryrelated studies.

Reducing hERG Toxicity Using hERG Classification Model and Fragment-growing Network

“…However, Chuipu Cai and co-workers chose a double threshold with large gap '≤10μM ≥80μM', with a dataset of only 4,447 data points, in comparison with our 9,708 data points. Similarly, although Li-li LIU et al 29 and Yiwei Wang et al 30 employed almost the same thresholds as us and obtained a quite good result, their data size is only 2,644, which is only about one third of ours.…”

“…Our work is compared with the previous studies in Table 6. From the 29 and Yiwei Wang et al 30 . However, Chuipu Cai and co-workers chose a double threshold with large gap '≤10μM ≥80μM', with a dataset of only 4,447 data points, in comparison with our 9,708 data points.…”

Machine learning models for classification tasks related to drug safety

“…Here, we have collected 15 works from the past five years that employ machine learning-based classification approaches to predict hERG inhibition [38][39][40][41][42][43][44][45][46][47][48][49][50][51]. All of these works apply training datasets of more than 1,000 molecules (and up to tens of thousands in some cases [47,48]), and an overall majority presents two-class (active vs. inactive) classification (with the notable example of the 2015 study of Braga et al, who have introduced a third class of "weak blockers") [38].…”

Artificial intelligence and machine learning‐aided drug discovery in central nervous system diseases: State‐of‐the‐arts and future directions

“…Here, capsule networks, 457 a next‐generation AI architecture where CNNs are encapsulated in an interconnected module, can provide a solution. As the first application of capsule networks to drug discovery, capsule networks showed excellent performance to predict the cardiotoxicity of compounds, which highlights their unique potential in drug discovery efforts 226 . Because of the modular representation of the CNNs, capsule networks can learn from heterogeneous data sets by preserving the hierarchical aspects of the data itself.…”

“…Moreover, the introduction of capsule networks, a new class of DNN architectures, has remarkably improved the ADME‐T prediction. To predict the cardiotoxicity of drugs, Wang et al 226 developed two capsule network architectures, including a convolution‐capsule network (Conv‐CapsNet) and a restricted Boltzmann machine‐capsule network (RBM‐CapsNet). Both models showed excellent performance with an accuracy of 91.8% for Conv‐CapsNet and 92.2% for RBM‐CapsNet.…”