Chemistry & ADMET · Swanson et al., Stanford · Chemprop, MIT
Chemprop D-MPNN — toxicity panel
Predicts small-molecule toxicity endpoints and structural alerts from chemical structures.
What Chemprop D-MPNN — toxicity panel is
Chemprop D-MPNN — toxicity panel is a molecular property prediction tool developed by Swanson et al. at Stanford and the Chemprop team at MIT. It represents the toxicity half of the ADMET-AI ensemble, focusing specifically on evaluating chemical safety liabilities in small molecules.
The model screens candidate compounds across multiple endpoints, including mutagenicity via the Ames test, cardiotoxicity via hERG inhibition, drug-induced liver injury, clinical trial toxicity, acute toxicity via LD50, and twelve Tox21 nuclear-receptor and stress-response pathways. It also identifies PAINS, Brenk, and NIH structural alerts.
By delivering toxicity predictions alongside percentile rankings contextualised against approved drugs, the tool enables computational chemists and medicinal chemists to triage molecular series and flag safety liabilities early in compound evaluation.
How it works
The model is built on a directed message-passing neural network (D-MPNN) architecture, a type of graph neural network tailored for molecular property prediction. It operates on molecular structures supplied as SMILES strings, computing graph representations along directed chemical bonds to predict multiple toxicity readouts concurrently.
Key facts
- Version
- 2.0.1
- Architecture
- Directed message-passing neural network (D-MPNN)
- Domain
- Chemistry
- Modalities
- Small molecule
- Inputs
- Ligand SMILES
- Outputs
- Toxicity endpoints with percentiles against approved drugs
What you can use it for
- Compound Series Liability Triaging
- Screens candidate small molecules to quickly flag mutagenic, cardiotoxic, or hepatotoxic risks during early drug discovery.
- Tox21 Pathway Profiling
- Evaluates compounds across all twelve Tox21 assays covering nuclear-receptor and stress-response pathways.
- Structural Alert Screening
- Identifies unwanted substructures by flagging PAINS, Brenk, and NIH alerts directly from input SMILES.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Runs on On-demand CPU
Runs on our servers with on-demand compute. A first run needs time to load the model; active capacity can be reused and scales down when idle.Run this model
Chemprop D-MPNN — toxicity panel
Strengths and limits
Good at
- Covers a wide range of toxicity endpoints including Ames, hERG, DILI, ClinTox, LD50, and the twelve Tox21 assays.
- Provides percentile comparisons directly against approved drugs to give biological context to prediction values.
- Incorporates standard chemical structural alert filters, including PAINS, Brenk, and NIH alerts.
- Runs efficiently on CPU hardware.
Watch out for
- Focuses exclusively on toxicity liabilities and structural alerts rather than broader absorption, distribution, or metabolism profiles.
- Accepts inputs only in the form of ligand SMILES strings.
Inputs and outputs
You provide
Ligand SMILES
Canonical SMILES for the small molecule.
You get back
Toxicity endpoints with percentile against approved drugs.
Licence
Commercial use permitted · MIT
The MIT licence allows use in commercial research and products.
Questions
Can I use Chemprop D-MPNN — toxicity panel commercially?
Commercial use permitted. The MIT licence allows use in commercial research and products.
How do I run Chemprop D-MPNN — toxicity panel without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers, typically in 900ms, and a run costs about 6 credits. Visitors without an account get 1,000 credits a month.
What inputs does the model require?
The model takes chemical structures provided as ligand SMILES strings.
What outputs does the model provide?
It outputs a table of predicted toxicity endpoints accompanied by percentile scores benchmarked against approved pharmaceuticals, alongside structural alert checks.
When should this model be used instead of ADMET-AI?
This model should be used when the focus is strictly on triaging toxicity endpoints and structural alerts. When absorption, distribution, and metabolism profiles are also required, the complete ADMET-AI model should be chosen.
Written from Chemprop D-MPNN — toxicity panel's own documentation. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.