ADMET Evaluation

The ADMET Evaluation function module comprises a series of high-quality prediction models trained by the Directed Message Passing Neural Network (DMPNN) framework. It enables users to conveniently and efficiently conduct calculations and predictions for 21 physicochemical properties, 19 medicinal chemistry properties, 34 ADME endpoints, 36 toxicity endpoints, and 8 toxicophore rules (751 substructures).

ADMET Screening

ADMET Screening is the batch mode of evaluation, designed for the prediction of molecular datasets. SMILES strings and SDF/TXT/CSV formatted files are supported molecular submission approaches. This module is suitable for the evaluation of empirically designed or visually screened molecules before chemical synthesis and biochemical assays, which enabling scientists to better focus their experiments on the most promising compounds.


Comprehensive coverage of ADMET endpoint data

ADMETlab 3.0 is more comprehensive and capable. In this version, the number of predictable endpoints has increased from 88 to 119, an addition of 30 compared to the previous version. The training model's dataset is 1.5 times larger than the previous version.

Robust and accurate multi-task DMPNN models

ADMETlab 3.0 utilizes the DMPNN framework, enabling message passing by fusing vectors of neighboring bonds in the molecular graph to enhance message aggregation and updating. Simultaneously, the integration of molecular graph vectors with molecular descriptors significantly improves the model's performance and robustness.

API integration and architecture upgrades

The introduction of an API interface caters to the growing demand for programmatic access to extensive data within ADMETlab 3.0. This API is designed for flexibility, enabling developers to leverage its functionality for broader applications or integrations. The existing website system architecture has undergone upgrades to enhance the user experience.

Incorporation of uncertainty evaluation

Providing uncertainty estimates for predictive results is essential to assess the accuracy of the prediction. It reflects the model's confidence in the prediction. Higher uncertainty may indicate that the model's prediction for the molecule is less reliable, while lower uncertainty implies greater confidence in the prediction.

Current Release: ADMETlab 3.0 | Last update: 2024-01-31 LICENSE