Multi-Modal Foundation · mist-models
MIST 1.8B · ClinTox
Predicts clinical trial toxicity and FDA approval status of chemical compounds from SMILES strings.
What MIST 1.8B · ClinTox is
MIST 1.8B · ClinTox is a molecular property prediction model developed by the Electrochemical Energy Group at the University of Michigan. It fine-tunes a 1.8-billion-parameter MIST (Molecular Insight SMILES Transformers) encoder on the ClinTox benchmark from MoleculeNet to assess pharmaceutical safety profiles.
The model evaluates molecular structures represented as SMILES (Simplified Molecular Input Line Entry System) strings, addressing two binary prediction endpoints: whether a candidate molecule failed clinical trials due to toxicity (CT_TOX) and whether it has attained FDA approval. It provides researchers with structural representations and classification outputs for screening therapeutic compounds.
How it works
The underlying architecture is a RoBERTa-PreLayerNorm encoder featuring 24 layers, 24 attention heads, a hidden size of 1536, and an intermediate size of 6144, supporting sequences up to 2048 tokens. The base encoder was pre-trained using a masked language modelling objective with cross-entropy loss on Kekulized SMILES strings from the Enamine REAL Space dataset, using 32 NVIDIA A100 GPUs and the FusedLAMB optimizer.
For property prediction, the encoder is paired with a task network comprising a two-layer multi-layer perceptron (MLP) with Gaussian Error Linear Units (GELU) activations and dropout. Encoder representations are pooled by taking the hidden state of the first token. The downstream head was fine-tuned on ClinTox using binary cross-entropy loss and the AdamW optimizer on a single NVIDIA A100 GPU to generate unnormalized classification logits.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder (24 layers, 24 heads) with a two-layer MLP classification head
- Parameter count
- 1.8 billion (encoder base)
- Pre-training dataset
- Enamine REAL Space
- Fine-tuning dataset
- MoleculeNet ClinTox (1,478 compounds)
- Context length
- 2048 tokens
- Serving memory requirement
- 24 GB GPU memory
What you can use it for
- Clinical trial toxicity screening
- Evaluates candidate small molecules to estimate the likelihood of clinical trial failure caused by toxicity (CT_TOX).
- FDA approval classification
- Assesses whether input chemical structures align with profiles of FDA-approved therapeutics.
- Molecular representation extraction
- Extracts pooled hidden state vectors from SMILES strings for downstream cheminformatics and machine learning 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 1.8B · ClinTox
Strengths and limits
Good at
- Pre-trained on the extensive Enamine REAL Space library of synthetically accessible compounds.
- Supports long molecular sequence inputs up to 2048 tokens.
- Jointly models both clinical trial toxicity failure and FDA approval status.
Watch out for
- Supervised fine-tuning is constrained to the ClinTox dataset size of 1,478 molecules.
- Requires chemical inputs to be formatted as Kekulized SMILES strings.
- Requires a Rust installation in the local environment to run the SMIRK tokenizer.
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 1.8B · 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 1.8B · 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 1× A100 16GB, typically in 12s, and a run costs about 70 credits. Visitors without an account get 1,000 credits a month.
What inputs does the model require?
The model takes molecular sequences formatted as Kekulized SMILES strings, tokenised using the SMIRK tokenizer library.
What tasks does the model predict?
It performs multi-label binary classification, outputting unnormalized logits for clinical trial toxicity (CT_TOX) and FDA approval status.
How was model performance evaluated during training?
Evaluation was performed on an 80/10/10 random data split using the Area Under the Receiver Operating Characteristic curve (AUROC) averaged across both ClinTox tasks.
What software prerequisites exist for tokenisation?
The SMIRK tokenizer requires a working Rust installation on the host system to build and execute.
How to cite it
Wadell, Alexius, Bhutani, Anoushka, Azumah, Victor, Ellis-Mohr, Austin R., Kelly, Celia, Zhao, Hancheng, Nayak, Anuj K., Hegazy, Kareem, Brace, Alexander, Lin, Hongyi, Emani, Murali, Vishwanath, Venkatram, Gering, Kevin, Alkan, Melisa, Gibbs, Tom, Wells, Jack, Varshney, Lav R., Ramsundar, Bharath, Duraisamy, Karthik, Mahoney, Michael W., Ramanathan, Arvind, and Viswanathan, Venkatasubramanian. Foundation Models for Discovery and Exploration in Chemical Space. arXiv:2510.18900 (2025). https://doi.org/10.48550/arXiv.2510.18900
Written from MIST 1.8B · 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.