Preprint
Machine Learning

ADMETlab 2.0: an integrated online platform for accurate and comprehensive predictions of ADMET properties

Guo‐Li Xiong(Central South University), Zhenhua Wu(Zhejiang University), Jiacai Yi(National University of Defense Technology), Li Fu(Central South University), Zhijiang Yang(Central South University), Chang‐Yu Hsieh(Tencent (China)), Mingzhu Yin(Central South University), Xiangxiang Zeng(Hunan University), Chengkun Wu(National University of Defense Technology), Aiping Lü(Hong Kong Baptist University), Xiang Chen(Central South University), Tingjun Hou(Zhejiang University), Dongsheng Cao(Central South University)
March 30, 2021Nucleic Acids Research2,757 citations
PDFDOI

2.8k

Citations

499

Influential Citations

Nucleic Acids Research

Venue

2021

Year

Abstract

Because undesirable pharmacokinetics and toxicity of candidate compounds are the main reasons for the failure of drug development, it has been widely recognized that absorption, distribution, metabolism, excretion and toxicity (ADMET) should be evaluated as early as possible. In silico ADMET evaluation models have been developed as an additional tool to assist medicinal chemists in the design and optimization of leads. Here, we announced the release of ADMETlab 2.0, a completely redesigned version of the widely used AMDETlab web server for the predictions of pharmacokinetics and toxicity properties of chemicals, of which the supported ADMET-related endpoints are approximately twice the number of the endpoints in the previous version, including 17 physicochemical properties, 13 medicinal chemistry properties, 23 ADME properties, 27 toxicity endpoints and 8 toxicophore rules (751 substructures). A multi-task graph attention framework was employed to develop the robust and accurate models in ADMETlab 2.0. The batch computation module was provided in response to numerous requests from users, and the representation of the results was further optimized. The ADMETlab 2.0 server is freely available, without registration, at https://admetmesh.scbdd.com/.

Analysis

Why This Paper Matters

ADMETlab 2.0 addresses a critical bottleneck in drug development: the high failure rate due to poor pharmacokinetics and toxicity. By providing a comprehensive, freely available online platform for predicting 88 ADMET-related endpoints, it enables medicinal chemists to evaluate and optimize lead compounds early in the pipeline. The integration of a multi-task graph attention framework represents a significant advancement over traditional single-task models, as it can capture complex relationships between different properties and improve prediction accuracy.

The platform's expansion to include 17 physicochemical properties, 13 medicinal chemistry properties, 23 ADME properties, 27 toxicity endpoints, and 8 toxicophore rules (751 substructures) makes it one of the most comprehensive tools available. The addition of a batch computation module addresses a common user request, facilitating high-throughput screening. This work is particularly relevant for AI practitioners working in drug discovery, as it demonstrates the practical application of graph neural networks to a real-world problem with significant impact.

Technical Contributions

  • Multi-task graph attention framework: The core innovation is the use of a graph attention mechanism that learns to weigh the importance of different atoms and bonds in a molecule for predicting multiple ADMET properties simultaneously. This allows the model to share representations across tasks, potentially improving generalization and data efficiency.
  • Expanded endpoint coverage: The platform supports 88 endpoints, roughly double the previous version, covering a wide range of physicochemical, medicinal chemistry, ADME, and toxicity properties. This comprehensive coverage is crucial for early-stage drug design.
  • Toxicophore rules: In addition to machine learning models, the platform includes 8 toxicophore rules based on 751 substructures, providing interpretable alerts for potential toxicity.
  • User-friendly features: The batch computation module and optimized result visualization make the tool practical for real-world use, lowering the barrier for adoption by medicinal chemists.

Results

The abstract does not provide specific quantitative results such as accuracy, AUC, or comparison benchmarks. However, the paper claims that the multi-task graph attention framework yields "robust and accurate models" and that the platform supports approximately twice the number of endpoints as the previous version. The high citation count (2757) suggests significant adoption and validation by the community. For concrete metrics, readers would need to consult the full paper.

Significance

ADMETlab 2.0 exemplifies how advanced machine learning techniques, specifically graph attention networks, can be deployed in a production web service to solve a critical problem in drug discovery. Its free availability without registration democratizes access to state-of-the-art ADMET prediction, potentially accelerating research in both academia and industry. The multi-task learning approach also provides a template for building comprehensive predictive platforms in other domains of cheminformatics and bioinformatics.

admet-predictiongraph-attentionmulti-task-learningdrug-discoveryweb-serverpharmacokineticstoxicity-prediction
Attention Mechanisms

Related Papers