Multi-Modal Foundation · mist-models

MIST 1.8B · μ (dipole)

A 1.8-billion-parameter foundation model predicting molecular dipole moments directly from SMILES strings.

What MIST 1.8B · μ (dipole) is

MIST 1.8B · μ (dipole) is a molecular property prediction model developed by the Electrochemical Energy Group at the University of Michigan, Ann Arbor. It pairs a large-scale transformer encoder with a dedicated regression head to predict the molecular dipole moment of chemical structures.

The dipole moment quantifies the net electrical polarity of a molecule caused by non-uniform charge distribution. Accurate estimation of this property is fundamental to understanding intermolecular interactions, dielectric behaviour, and solvation dynamics across chemical spaces, including candidate electrolytes, pharmaceuticals, and functional materials.

How it works

The core encoder is based on a RoBERTa-PreLayerNorm architecture comprising 24 layers, 18 attention heads, a hidden size of 2304, and an intermediate dimension of 9216, capable of handling sequence lengths up to 2048 tokens. The model was initially pre-trained using a masked language modelling objective on Kekulized SMILES strings from the Enamine REAL Space chemical database, using cross-entropy loss and the FusedLAMB optimizer.

For property estimation, the pre-trained encoder is paired with a task network consisting of a two-layer multi-layer perceptron with Gaussian Error Linear Unit (GELU) activations and dropout. Representations are extracted by pooling the hidden state of the first token. The network was fine-tuned on the QM9 dipole moment dataset from MoleculeNet using mean squared error loss and the AdamW optimizer.

Key facts

Architecture: RoBERTa-PreLayerNorm encoder with a 2-layer MLP regression head
Parameter count: 1.8 billion
Pre-training data: Enamine REAL Space
Fine-tuning data: QM9 Dipole Moment (MoleculeNet)
Context length: 2048 tokens
Input format: Kekulized SMILES strings
Output format: Dipole moment in Debye (D)

What you can use it for

Molecular Polarity Profiling: Estimating molecular dipole moments in Debye directly from SMILES strings to evaluate electrostatic asymmetry without running quantum mechanical simulations.
Electrolyte and Solvent Screening: Assessing polarity trends across candidate chemical structures to guide formulation design for electrochemical and battery systems.
High-Throughput Property Prediction: Predicting dipole characteristics across batches of candidate compounds during early-stage virtual screening 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 · μ (dipole)

Strengths and limits

Good at

Predicts quantum-mechanical dipole values directly from 1D SMILES representations without requiring 3D conformer generation.
Leverages representations learned from the synthetic compound collection in Enamine REAL Space.
Accepts inputs of up to 2048 tokens, enabling support for large molecular SMILES sequences.

Watch out for

Models were trained on Kekulized SMILES representations and may yield unreliable outputs if given alternative or non-Kekulized formats.
Fine-tuning was conducted on the small-molecule QM9 dataset, which may limit generalisation to large biomolecules or atypical chemical frameworks.
Tokenisation depends on SMIRK tokenizers, requiring a working Rust environment for execution.

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 · μ (dipole) 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 · μ (dipole) 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 units are the predictions reported in?

The model outputs dipole moment predictions in Debye (D).

What specific molecular representation must be supplied as input?

Inputs must be provided as SMILES strings, specifically in Kekulized form.

Which dataset was used to train the dipole prediction head?

The dipole head was fine-tuned on the QM9 dataset taken from the MoleculeNet benchmark.

How are molecular sequence tokens aggregated for regression?

The network extracts the final hidden state of the first token in the sequence as the pooled representation, which is then passed 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 1.8B · μ (dipole)'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.