Multi-Modal Foundation · mist-models
MIST 28M · ToxCast
A 28-million parameter transformer predicting 617 toxicity assay endpoints directly from chemical SMILES strings.
What MIST 28M · ToxCast is
MIST 28M · ToxCast 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, designed to provide foundation representations for chemical exploration and property estimation.
This specific variant is fine-tuned to predict multi-label toxicity outcomes across 617 high-throughput screening assays from the United States Environmental Protection Agency (EPA) ToxCast project, available through the MoleculeNet benchmark. Given molecular SMILES strings, it outputs predictions across diverse in vitro biological pathways and toxicity endpoints.
By evaluating toxicology endpoints computationally, the model allows researchers in chemistry, pharmacology, and environmental science to assess the potential hazardous activity of small molecules without requiring immediate laboratory screening.
How it works
The model employs a RoBERTa-PreLayerNorm encoder with 8 layers, an intermediate size of 2048, a hidden dimension of 512, 8 attention heads, and a context window of 2048 tokens. It was initially pre-trained using a masked language modelling (MLM) cross-entropy objective on Kekulized SMILES strings from the Enamine REAL Space dataset.
For downstream toxicity prediction, the pre-trained encoder (mist-28M-ti624ev1) is coupled to a task network comprising a two-layer multi-layer perceptron (MLP) with Gaussian Error Linear Units (GELU) activations and dropout. Input SMILES strings are processed by taking the hidden state of the first token as the sequence embedding, and the model is fine-tuned via binary cross-entropy loss for multi-label classification.
Key facts
- Model architecture
- RoBERTa-PreLayerNorm encoder with a 2-layer MLP head
- Parameters
- 28 million
- Maximum context length
- 2048 tokens
- Pre-training data
- Enamine REAL Space SMILES strings
- Fine-tuning data
- MoleculeNet ToxCast dataset (over 8,000 compounds across 617 assays)
- Primary input format
- Kekulized SMILES strings
What you can use it for
- High-throughput toxicity screening
- Predicting unnormalised classification logits across 617 ToxCast in vitro screening assays for compound libraries.
- In vitro pathway assessment
- Evaluating potential interactions between candidate molecules and specific biological pathways defined in the EPA ToxCast programme.
- Molecular representation extraction
- Generating single-vector pooled embeddings from molecular SMILES for downstream analytical tasks.
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 · ToxCast
Strengths and limits
Good at
- Simultaneously predicts outcomes for 617 distinct toxicity assay endpoints in a multi-task framework.
- Pre-trained on Enamine REAL Space, capturing synthetically accessible chemical space.
- Supports long sequence lengths up to 2048 tokens for complex molecular inputs.
Watch out for
- Requires Kekulized SMILES strings as inputs rather than alternative molecular representations.
- Tokenization depends on the external SMIRK tokenizer, which requires a Rust installation.
- Outputs unnormalised logits that must be calibrated or interpreted appropriately for multi-label binary tasks.
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 · ToxCast 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 · ToxCast 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 inputs does the model accept?
The model accepts chemical SMILES strings, typically formatted in Kekulized form.
What does the model output during inference?
It returns unnormalised logits for multi-label binary classification across 617 ToxCast assay endpoints, or pooled sequence embeddings when used for feature extraction.
How were the ToxCast training and evaluation sets split?
The ToxCast dataset was divided into training, validation, and test subsets using an 80/10/10 random split.
Which metric is used to benchmark model performance?
Performance is measured using the Area Under the Receiver Operating Characteristic curve (AUROC) averaged across all 617 assay tasks.
How to cite it
Wadell, A., Bhutani, A., Azumah, V., Ellis-Mohr, A. R., Kelly, C., Zhao, H., Nayak, A. K., Hegazy, K., Brace, A., Lin, H., Emani, M., Vishwanath, V., Gering, K., Alkan, M., Gibbs, T., Wells, J., Varshney, L. R., Ramsundar, B., Duraisamy, K., Mahoney, M. W., Ramanathan, A., & Viswanathan, V. (2025). Foundation Models for Discovery and Exploration in Chemical Space. arXiv preprint arXiv:2510.18900.
Written from MIST 28M · ToxCast'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.