ADMETPred: a high-throughput ADMET prediction platform integrating multi-model algorithms and interpretable...

Area of Science:

• Computational Chemistry
• Drug Discovery
• Artificial Intelligence

Background:

• Accurate prediction of ADMET properties is crucial for efficient drug discovery.
• Current computational tools face limitations in throughput, interpretability, and flexibility.
• AI offers transformative potential but requires advanced modeling approaches.

Purpose of the Study:

• To develop an innovative platform, ADMETPred, for rapid, accurate, and comprehensive ADMET profiling.
• To overcome limitations of existing computational tools in drug discovery.
• To provide actionable insights for structural optimization in drug development.

Main Methods:

• Integration of machine learning and graph neural networks.
• Training on a curated dataset of 120,616 compounds using 189 models (LightGBM, XGBoost, Random Forest, GAT).
• Development of a high-throughput, parallelized architecture with customizable workflows and an interpretable substructure highlighting module.

Main Results:

• ADMETPred demonstrates superior predictive accuracy for 27 drug pharmacokinetic, metabolism, and toxicity endpoints compared to existing tools.
• The platform offers high-throughput batch processing and enhanced prediction flexibility.
• Case studies confirmed alignment with experimental and clinical evidence in drug surveillance, toxicity screening, and safety assessment.

Conclusions:

• ADMETPred provides a practical and accessible resource for enhancing early-stage drug development.
• The platform lowers usage barriers while delivering reliable ADMET profiling.
• ADMETPred facilitates informed decision-making through interpretable predictions and structural insights.