Recent Advances in Toxicity Prediction: Applications of Deep Graph Learning

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Drug development is costly and time-consuming, necessitating accurate toxicity prediction for safety and efficacy.
• Deep graph learning (DGL) offers computational power and cost-efficiency for predicting drug toxicity.
• Existing methods require comprehensive understanding of DGL components and applications in toxicology.

Purpose of the Study:

• To bridge fundamental knowledge with advanced deep graph learning methods for drug toxicity prediction.
• To provide a comprehensive overview of DGL components, including molecular descriptors, representations, metrics, validation, and datasets.
• To review representative DGL studies and methods, focusing on GNN architectures and graph pretrained models.

Main Methods:

• Summarizing essential components of DGL models for toxicity prediction.
• Analyzing various graph-based molecular representations.
• Introducing representative studies from the perspective of GNN architectures and graph pretrained models.

Main Results:

• Deep graph models demonstrate superior accuracy and efficiency compared to other approaches.
• DGL models offer more intuitive insights, enhancing model interpretation and generalization.
• Graph pretrained models show promise in extracting features from large unlabeled molecular data for improved toxicity prediction.

Conclusions:

• Deep graph learning is a powerful and efficient tool for advancing drug toxicity prediction.
• Graph pretrained models are emerging as key enablers for enhancing downstream toxicity prediction tasks.
• This review serves as a guide for researchers entering the field of DGL for drug toxicity prediction.