Use ADMET Online
Commercially Available ADMET No-Code Web Server
ADMET-AI: Fast Machine Learning ADMET Property Prediction for Drug Candidates
ADMET-AI is a machine learning platform for predicting absorption, distribution, metabolism, excretion and toxicity (ADMET) properties of small molecules and evaluating large chemical libraries. It was developed by Kyle Swanson, James Zou and colleagues at Stanford University and Greenstone Biosciences, and published as an Applications Note in Bioinformatics in 2024. The article is open access under CC BY 4.0, and the code is open source on GitHub.
Large chemical libraries, high-throughput docking and generative AI have greatly expanded the number of candidate small molecules in drug discovery. According to the authors, many of these hits lack favorable drug-like properties, so they need to be screened for ADMET properties before experimental validation. ADMET-AI is built to make that screening fast enough for large libraries.
Property | Detail |
|---|---|
Task | Prediction of 41 ADMET endpoints from molecular structure (SMILES) |
Model | Chemprop-RDKit: a graph neural network augmented with 200 RDKit physicochemical features |
Training data | 41 ADMET datasets from the Therapeutics Data Commons (10 regression, 31 classification) |
Reference set | 2,579 approved drugs from DrugBank v5.1.10, used to report percentiles |
Web server limit | Up to 1,000 molecules per request |
Local package | Python command line tool and module with no molecule limit |
How ADMET-AI Predicts ADMET Properties
Graph neural network with chemical descriptors: the graph representation of each molecule is concatenated with 200 physicochemical features and passed through feed-forward layers.
Ensembling: one model is trained per dataset for each of five train, validation and test splits, and inference averages the five-model ensemble.
Multi-task models: two multi-task models, one for regression and one for classification, cover all 41 properties. The authors report they perform comparably to single-task models and are much faster, and they power the web server.
DrugBank percentiles: each prediction is converted to a percentile relative to approved drugs, which can be filtered by Anatomical Therapeutic Chemical (ATC) code.
ADMET-AI Benchmark Results and Speed
Evaluation | Result |
|---|---|
TDC ADMET Leaderboard (22 datasets) | Best average rank among compared models |
Regression datasets | R-squared above 0.6 for 5 of 10 |
Classification datasets | AUROC above 0.85 for 20 of 31 |
Web server speed (1,000 molecules) | About 45% faster than the next-fastest public ADMET web server |
1 million molecules, local run | 3.1 hours with 32 CPU cores and a GPU, vs. an estimated 28.1 hours on the web server |
The authors benchmarked speed against SwissADME, ADMETlab 2.0, pkCSM, vNN-ADMET, ADMETboost and PreADMET, noting that some of these cap input size, for example pkCSM at 100 molecules and SwissADME at 200.
Web Server Outputs
The web server accepts SMILES as text, CSV upload or a drawn structure. It displays the 41 ADMET predictions plus eight RDKit physicochemical properties for up to 25 molecules, with the rest available for download, and includes a radar plot of drug-likeness and a scatter plot comparing molecules to the reference set.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with ADMET-AI on Tamarind Bio
Library triage: screen large chemical libraries and remove compounds with poor predicted ADMET profiles before costly experiments.
Docking and generative follow-up: evaluate hits from high-throughput docking or generative AI for drug-likeness.
Context against approved drugs: compare each compound with DrugBank approved drugs through percentile scores.
Key safety and exposure endpoints: review summary properties such as hERG, blood-brain barrier penetration, solubility, oral bioavailability and toxicity.
How to Use ADMET-AI on Tamarind Bio
Log in: sign in at tamarind.bio and open the ADMET tool.
Provide molecules: enter SMILES strings or upload a file of SMILES, as the tool accepts.
Choose a reference subset: if exposed, pick the DrugBank reference set or an ATC-code subset for percentile context.
Run the job: submit it and wait for predictions.
Download outputs: retrieve the table of ADMET predictions and physicochemical properties.
Interpret percentiles: compare each property with approved drugs to see where a compound stands.
Prioritize: shortlist compounds with favorable profiles for the next stage of the pipeline.
Things to Keep in Mind
The web server in the paper is capped at 1,000 molecules per request, so larger jobs need a local installation there.
The DrugBank reference set contains only approved drugs, so percentiles are relative to that population.
Some timing comparisons in the paper are extrapolated rather than measured.
The paper has no dedicated limitations section; predictions are computational estimates and need experimental confirmation.
Source: Swanson K, Walther P, Leitz J, Mukherjee S, Wu JC, Shivnaraine RV, Zou J. ADMET-AI: a machine learning ADMET platform for evaluation of large-scale chemical libraries. Bioinformatics 40(7):btae416, 2024. DOI: 10.1093/bioinformatics/btae416. Code: github.com/swansonk14/admet_ai.