Recent Advances in Toxicity Prediction: Applications of Deep Graph Learning

The development of new drugs is time-consuming and expensive,andas such, accurately predicting the potential toxicity of a drug candidateis crucial in ensuring its safety and efficacy. Recently, deep graphlearning has become prevalent in this field due to its computationalpower and cost efficiency. Many novel deep graph learning methodsaid toxicity prediction and further prompt drug development. Thisreview aims to connect fundamental knowledge with burgeoning deepgraph learning methods. We first summarize the essential componentsof deep graph learning models for toxicity prediction, including moleculardescriptors, molecular representations, evaluation metrics, validationmethods, and data sets. Furthermore, based on various graph-relatedrepresentations of molecules, we introduce several representativestudies and methods for toxicity prediction from the perspective ofGNN architectures and graph pretrained models. Compared to other typesof models, deep graph models not only advance in higher accuracy andefficiency but also provide more intuitive insights, which is significantin the development of model interpretation and generalization ability.The graph pretrained models are emerging as they can extract prominentfeatures from large-scale unlabeled molecular graph data and improvethe performance of downstream toxicity prediction tasks. We hope thissurvey can serve as a handbook for individuals interested in exploringdeep graph learning for toxicity prediction.