Multi-Modal Foundation · mist-models
MIST 1.8B · ⟨R²⟩
Predicts molecular electronic spatial extent from SMILES strings using a 1.8-billion-parameter transformer encoder.
What MIST 1.8B · ⟨R²⟩ is
MIST 1.8B · ⟨R²⟩ 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, which applies self-supervised transformer architectures to chemical representations. This variant adapts a 1.8-billion-parameter foundation encoder specifically for estimating the electronic spatial extent of small molecules.
Electronic spatial extent, denoted as ⟨R²⟩, describes the spatial distribution of a molecule's electron density and reflects the physical dimensions and spread of its electron cloud. Calculating this property traditionally requires quantum chemical calculations; this model instead performs rapid regression directly from chemical string representations derived from molecular structures.
How it works
The underlying network is a RoBERTa-PreLayerNorm encoder containing 24 layers, 24 attention heads, a hidden dimension of 1536, an intermediate size of 6144, and a maximum sequence context of 2048 tokens. It was pre-trained on Kekulized SMILES strings from the Enamine REAL Space database using a masked language modelling objective with cross-entropy loss and the FusedLAMB optimizer across 32 NVIDIA A100 GPUs.
For property prediction, the encoder is paired with a task head consisting of a two-layer multi-layer perceptron with Gaussian Error Linear Unit (GELU) activations and dropout. The representation is formed by pooling the hidden state of the first token. The network was fine-tuned on the QM9 dataset using mean squared error loss and the AdamW optimizer on a single NVIDIA A100 GPU to predict ⟨R²⟩ in units of Bohr².
Key facts
- Architecture
- RoBERTa-PreLayerNorm with a two-layer MLP prediction head
- Parameter count
- 1.8 billion
- Pre-training data
- Enamine REAL Space
- Fine-tuning data
- QM9 (MoleculeNet)
- Target property
- Electronic spatial extent (⟨R²⟩) in Bohr²
- Context length
- 2048 tokens
What you can use it for
- Electronic Spatial Extent Estimation
- Predicting the electronic spatial extent (⟨R²⟩) of molecules directly from SMILES strings without running density functional theory simulations.
- Molecular Cloud Profiling
- Assessing the overall physical size and electron cloud spread across candidate molecular libraries.
- Molecular Representation Extraction
- Generating fixed-length feature vectors from chemical SMILES inputs via the transformer encoder's pooling mechanism.
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 · ⟨R²⟩
Strengths and limits
Good at
- Pre-trained on the vast, synthetically accessible chemical space of the Enamine REAL Space database.
- Supports context windows up to 2048 tokens for handling extended molecular string representations.
- Bypasses the requirement for 3D atomic coordinates by predicting quantum chemical properties straight from 1D SMILES strings.
Watch out for
- Input preparation requires Kekulized SMILES strings and tokenisation via the SMIRK library, which depends on a Rust installation.
- Task adaptation is confined to electronic spatial extent as calibrated on the small-molecule QM9 dataset.
- Weights are supplied for research purposes only without formal performance guarantees.
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 · ⟨R²⟩ 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 · ⟨R²⟩ 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 this model require?
The model expects molecular SMILES strings that have been Kekulized, tokenised using the SMIRK tokenizer.
What unit is the predicted electronic spatial extent reported in?
The regression head outputs predictions in atomic units of Bohr squared (Bohr²).
How was the model evaluated during development?
It was evaluated using Mean Absolute Error (MAE) on a random 80/10/10 split of the QM9 dataset.
What base checkpoint was used for fine-tuning?
The model was fine-tuned from the pre-trained base encoder mist-1.8B-dh61satt.
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 1.8B · ⟨R²⟩'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.