Multi-Modal Foundation · mist-models
MIST 1.8B · U298
Predicts molecular internal energy at 298.15 K from SMILES representations using a 1.8-billion-parameter transformer.
What MIST 1.8B · U298 is
MIST 1.8B · U298 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 and is specifically fine-tuned to predict the internal energy at 298.15 K (U298) of chemical structures.
Internal energy at 298.15 K is a quantum chemical benchmark property from the QM9 dataset within MoleculeNet. It represents the total room-temperature internal energy of a molecule, incorporating thermal contributions from molecular vibrations, rotations, and translations. By estimating this property directly from chemical strings, the model assists computational chemists in screening and prioritising compounds across chemical space.
How it works
The model is based on a RoBERTa-PreLayerNorm encoder consisting of 24 layers, a hidden dimension of 2,304, an intermediate dimension of 9,216, and 18 attention heads, supporting sequence lengths up to 2,048 tokens. The underlying encoder was pre-trained on Kekulized SMILES strings from the Enamine REAL Space database using a masked language modelling objective and cross-entropy loss.
For property prediction, the pre-trained encoder is coupled with a task network consisting of a two-layer multi-layer perceptron with Gaussian Error Linear Unit (GELU) activations and dropout. Hidden states are pooled using the representation of the first sequence token, and the network is fine-tuned on QM9 U298 values using mean squared error loss.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder with a two-layer MLP prediction head
- Parameter count
- 1.8 billion
- Maximum sequence length
- 2,048 tokens
- Pre-training dataset
- Enamine REAL Space
- Fine-tuning dataset
- QM9 (MoleculeNet)
- Target property
- Internal energy at 298.15 K (U298) in Hartree
What you can use it for
- Room-temperature internal energy estimation
- Predicts U298 directly from Kekulized molecular SMILES strings in Hartree without running density functional theory calculations.
- Thermodynamic screening of compound libraries
- Screens candidate molecules such as electrolytes or drug leads to assess stability and internal energy profiles across large chemical libraries.
- Molecular representation extraction
- Extracts pooled transformer embedding vectors from SMILES inputs for downstream classification or clustering pipelines.
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 · U298
Strengths and limits
Good at
- Pre-trained on the vast, synthetically accessible chemical space of the Enamine REAL Space database.
- Captures thermal contributions from vibrations, rotations, and translations within room-temperature internal energy estimates.
- Supports long chemical inputs up to a context window of 2,048 tokens.
Watch out for
- Fine-tuned on the QM9 dataset, which is restricted to small organic molecules and may not transfer well to larger or diverse structures.
- Requires molecular structures to be formatted as SMILES, primarily Kekulized SMILES.
- Requires the SMIRK tokenizer, which relies on a Rust installation in the user environment.
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 · U298 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 · U298 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 input format does the model take?
The model expects molecular SMILES strings, specifically Kekulized SMILES, tokenised using the SMIRK fast tokeniser.
What units are the predictions reported in?
Predictions for internal energy at 298.15 K are generated in Hartree.
What does the U298 property measure?
It measures the total internal energy of a molecule at room temperature (298.15 K), including vibrational, rotational, and translational thermal energy contributions.
Can this model be used as an embedding generator?
Yes, the model can be run as a feature-extraction pipeline to produce vector embeddings representing the input molecules.
How to cite it
Wadell, Alexius and Bhutani, Anoushka and Azumah, Victor and Ellis-Mohr, Austin R. and Kelly, Celia and Zhao, Hancheng and Nayak, Anuj K. and Hegazy, Kareem and Brace, Alexander and Lin, Hongyi and Emani, Murali and Vishwanath, Venkatram and Gering, Kevin and Alkan, Melisa and Gibbs, Tom and Wells, Jack and Varshney, Lav R. and Ramsundar, Bharath and Duraisamy, Karthik and Mahoney, Michael W. and Ramanathan, Arvind and Viswanathan, Venkatasubramanian, 'Foundation Models for Discovery and Exploration in Chemical Space', arXiv:2510.18900, 2025.
Written from MIST 1.8B · U298'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.