Multi-Modal Foundation · mist-models
MIST 28M · ClinTox
Predicts clinical toxicity and FDA approval status from molecular SMILES strings for drug candidates.
What MIST 28M · ClinTox is
MIST 28M · ClinTox is a molecular property prediction model developed by the Electrochemical Energy Group at the University of Michigan, Ann Arbor. It belongs to the Molecular Insight SMILES Transformers (MIST) family of foundation models, designed to evaluate and navigate chemical space using sequential molecular representations.
The model addresses clinical attrition by identifying safety risks early in drug development. By learning from clinical outcomes, it classifies chemical compounds based on whether they failed clinical trials due to toxicity and whether they achieved regulatory approval.
How it works
The architecture consists of an 8-layer RoBERTa-PreLayerNorm encoder with a hidden dimension of 512, an intermediate size of 2048, 8 attention heads, and a context window of 2048 tokens. The base encoder was pre-trained using a masked language modelling objective on Kekulized SMILES strings from the Enamine REAL Space collection, optimized with FusedLAMB.
For clinical toxicity classification, the encoder is coupled to a task network comprising a two-layer multi-layer perceptron with Gaussian Error Linear Unit (GELU) activations and dropout. Hidden states are pooled from the first sequence token. The model is fine-tuned using binary cross-entropy loss and AdamW on the ClinTox benchmark, which contains 1,478 compounds annotated for FDA approval status and clinical trial toxicity withdrawal.
Key facts
- Architecture
- RoBERTa-PreLayerNorm with a two-layer MLP classification head
- Parameter count
- 28 million
- Pre-training dataset
- Enamine REAL Space
- Fine-tuning dataset
- MoleculeNet ClinTox (1,478 compounds)
- Maximum sequence length
- 2048 tokens
- Input format
- Kekulized SMILES strings
- Output format
- Unnormalised binary classification logits across 2 tasks
What you can use it for
- Clinical Trial Toxicity Screening
- Screens prospective small-molecule drug candidates to identify potential toxicity failures prior to clinical stages.
- FDA Approval Status Prediction
- Assesses the likelihood that a compound matches the chemical and safety profiles of approved therapeutics.
- Molecular Feature Extraction
- Generates dense embedding vectors from SMILES inputs using the fine-tuned encoder for clustering or downstream analysis.
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 GPU
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
MIST 28M · ClinTox
Strengths and limits
Good at
- Pre-trained on the Enamine REAL Space database, providing strong familiarity with synthetically accessible chemical space.
- Predicts both clinical trial failure and regulatory approval simultaneously within a single multi-label framework.
- Supports long SMILES inputs up to a context length of 2048 tokens.
Watch out for
- Fine-tuning is restricted to the relatively small ClinTox dataset of 1,478 chemical compounds.
- Relies entirely on 1D SMILES sequence inputs, lacking explicit three-dimensional stereochemical or spatial coordinates.
- Outputs raw, unnormalised logits rather than calibrated probabilities of clinical failure.
Inputs and outputs
You provide
Sequence or text
A protein or nucleotide sequence, or free text — whichever this model was trained on.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use MIST 28M · ClinTox commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
How do I run MIST 28M · ClinTox 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 on CPU only, typically in 12s, and a run costs about 28 credits. Visitors without an account get 1,000 credits a month.
What input formats are required?
The model accepts molecular SMILES strings, specifically requiring Kekulized SMILES tokenised with the SMIRK tokeniser.
What are the model's exact prediction targets?
It generates multi-label classification outputs corresponding to two specific tasks: clinical trial toxicity failure (CT_TOX) and FDA approval status.
How was the model evaluated during development?
It was tested using a random 80/10/10 split of the ClinTox dataset and assessed using the Area Under the Receiver Operating Characteristic curve (AUROC) averaged across both prediction tasks.
Can the model output molecular representations instead of predictions?
Yes, it can produce mean-pooled embedding vectors or first-token hidden states representing the input sequence.
How to cite it
Alexius Wadell, Anoushka Bhutani, Victor Azumah, Austin R. Ellis-Mohr, Celia Kelly, Hancheng Zhao, Anuj K. Nayak, Kareem Hegazy, Alexander Brace, Hongyi Lin, Murali Emani, Venkatram Vishwanath, Kevin Gering, Melisa Alkan, Tom Gibbs, Jack Wells, Lav R. Varshney, Bharath Ramsundar, Karthik Duraisamy, Michael W. Mahoney, Arvind Ramanathan, and Venkatasubramanian Viswanathan. Foundation Models for Discovery and Exploration in Chemical Space. arXiv:2510.18900 [physics], 2025. https://doi.org/10.48550/arXiv.2510.18900
Written from MIST 28M · ClinTox's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.