Multi-Modal Foundation · mist-models
MIST 26.9M · flash point
A molecular transformer predicting flash-point temperature from Kekulized SMILES strings.
What MIST 26.9M · flash point is
MIST 26.9M · flash point is a molecular property prediction model developed by the Electrochemical Energy Group at the University of Michigan. It builds on the Molecular Insight SMILES Transformers (MIST) family of foundation models, designed to explore chemical space and predict thermodynamic properties relevant to chemistry and energy engineering.
The model addresses the prediction of flash point—the minimum temperature at which a chemical compound's vapours ignite, which serves as a critical indicator of flammability. Obtaining experimental flash points can be resource-intensive, and this model enables direct estimation from molecular chemical structure strings.
Trained on an encoder-plus-task-network framework, it ingests Kekulized SMILES representations and outputs a continuous regression prediction for flash point in Kelvin (K). It can also function in feature-extraction pipelines to yield pooled molecular vector embeddings.
How it works
The architecture uses a RoBERTa-PreLayerNorm encoder with 8 layers, a hidden dimension of 512, an intermediate dimension of 2048, 8 attention heads, and a maximum sequence length of 2048 tokens. For property prediction, a two-layer multi-layer perceptron (MLP) task head with Gaussian Error Linear Units (GELU) and dropout is attached. The task head takes the pooled hidden state corresponding to the first token of the sequence.
Pretraining was conducted on molecular SMILES strings from the Enamine REAL Space database using a Masked Language Modeling (MLM) cross-entropy objective and the FusedLAMB optimizer. The model was subsequently fine-tuned for flash-point regression using Mean Squared Error (MSE) loss and the AdamW optimizer on 10,090 literature entries from the Characteristic Temperatures dataset collected by Shang Zhu et al., restricted to compounds with fewer than 20 non-hydrogen heavy atoms and zero radical electrons.
Key facts
- Model architecture
- RoBERTa-PreLayerNorm (8 layers, hidden size 512, 8 heads) with 2-layer MLP
- Parameter count
- 26.9M (base encoder derived from mist-28M-ti624ev1)
- Pretraining dataset
- Enamine REAL Space
- Fine-tuning dataset
- Flash Point dataset (Shang Zhu et al. Characteristic Temperatures dataset, 10,090 entries)
- Input format
- Kekulized SMILES strings (processed with SmirkTokenizerFast)
- Maximum sequence length
- 2048 tokens
- Primary output
- Flash point temperature in Kelvin (K)
What you can use it for
- Flammability screening for organic molecules
- Estimate the flash point (in Kelvin) directly from Kekulized SMILES strings to evaluate flammability hazards during early-stage chemical formulation.
- Electrolyte and solvent safety evaluation
- Screen candidate battery electrolyte solvents or additives under 20 heavy atoms to prioritize thermally stable and fire-resistant compounds.
- Molecular representation extraction
- Extract mean-pooled or first-token embedding vectors from the fine-tuned RoBERTa encoder for clustering or training secondary chemical property models.
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 26.9M · flash point
Strengths and limits
Good at
- Pretrained on the Enamine REAL Space library, covering chemically synthesizable regions of molecular space.
- Accepts sequence lengths up to 2048 tokens, enabling support for long Kekulized SMILES strings.
- Combines a transformer encoder with a lightweight two-layer MLP head for fast downstream flash-point inference.
Watch out for
- Fine-tuning was restricted to small molecules with fewer than 20 non-hydrogen heavy atoms, limiting reliability on larger structures.
- Excluded compounds with radical electrons during fine-tuning, making it unsuitable for open-shell or radical species.
- Requires Kekulized SMILES strings and dependencies such as Rust for the SMIRK tokenizer to process inputs properly.
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 26.9M · flash point 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 26.9M · flash point 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 units are used for the flash point output?
The model outputs regression predictions for flash point temperature in Kelvin (K).
What input format does the model expect?
The model expects SMILES strings of molecules, which should be Kekulized. Input tokenization uses SmirkTokenizerFast.
Can this model be applied to macromolecules or polymers?
No. The fine-tuning data was strictly filtered to molecules containing fewer than 20 heavy atoms (excluding hydrogen), so predictions on macromolecules fall outside the training domain.
How were the hidden states pooled for property prediction?
The task network takes the encoder's hidden state of the first sequence token as the pooled representation before passing it to the two-layer MLP.
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:2510.18900 [physics]. https://doi.org/10.48550/arXiv.2510.18900
Written from MIST 26.9M · flash point'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.