Multi-Modal Foundation · mist-models
MIST 1.8B · α (polarizability)
A 1.8-billion parameter molecular transformer predicting isotropic polarizability directly from chemical SMILES.
What MIST 1.8B · α (polarizability) is
MIST 1.8B · α (polarizability) is a molecular foundation model fine-tuned to predict the isotropic polarizability (α) of chemical compounds. Developed by the Electrochemical Energy Group at the University of Michigan, Ann Arbor, the model builds upon the Molecular Insight SMILES Transformers (MIST) family designed for molecular property prediction.
Polarizability describes how easily a molecule's electron cloud is distorted by an external electric field, serving as a critical physical parameter for understanding intermolecular forces, chemical reactivity, and optical characteristics. By learning directly from chemical notations, the model provides numerical estimates of isotropic polarizability without requiring computationally demanding quantum chemical simulations.
How it works
The underlying foundation model relies on a RoBERTa-PreLayerNorm encoder architecture consisting of 24 layers, 24 attention heads, a hidden dimension of 1536, an intermediate dimension of 6144, and a context window of up to 2048 tokens. The encoder was initially pre-trained using a masked language modelling (MLM) objective on Kekulized SMILES strings from the Enamine REAL Space dataset. Pre-training was optimised with Cross-Entropy Loss using the FusedLAMB optimiser.
For downstream property prediction, the pre-trained encoder is coupled to a task-specific network comprising a two-layer multi-layer perceptron (MLP) with Gaussian Error Linear Units (GELU) activations and dropout. Sequence representation is obtained by pooling the final hidden state of the first token. The model was fine-tuned on the QM9 isotropic polarizability benchmark using Mean Squared Error (MSE) loss and the AdamW optimiser, outputting values in units of cubic Bohr (Bohr³).
Key facts
- Base architecture
- RoBERTa-PreLayerNorm encoder with a 2-layer MLP head
- Parameter count
- 1.8 billion
- Context length
- 2048 tokens
- Pre-training data
- Enamine REAL Space SMILES
- Fine-tuning data
- QM9 Isotropic Polarizability dataset (MoleculeNet)
- Input format
- Kekulized SMILES strings
- Output property
- Isotropic polarizability (Bohr³)
What you can use it for
- Predicting Molecular Polarizability
- Estimate the isotropic polarizability (α) in Bohr³ directly from molecular SMILES strings, useful for evaluating interactions in candidate compounds.
- Optical and Electronic Property Screening
- Assess electronic deformability and optical response tendencies across candidate molecules for electrolyte, fragrance, or drug design tasks.
- Molecular Representation Extraction
- Extract mean-pooled sequence embeddings from the fine-tuned RoBERTa-PreLayerNorm encoder for downstream chemoinformatic clustering or transfer learning.
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 · α (polarizability)
Strengths and limits
Good at
- Pre-trained on the vast, synthetically accessible chemical space of the Enamine REAL Space dataset.
- Directly outputs isotropic polarizability predictions in standard physical units (Bohr³).
- Employs a RoBERTa-PreLayerNorm architecture with a long context window of up to 2048 tokens.
Watch out for
- Requires input molecules to be provided as Kekulized SMILES strings.
- Requires Rust to be installed locally to run its SMIRK tokenizer dependencies.
- Evaluation on QM9 relies on an 80/10/10 random split rather than a scaffold-based split, which may affect generalisation 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 · α (polarizability) 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 · α (polarizability) 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 polarizability predictions reported in?
The model outputs regression predictions for isotropic polarizability in cubic Bohr (Bohr³).
What format must input molecules take?
The model accepts chemical SMILES strings, specifically expecting Kekulized SMILES.
What software dependencies are required for tokenisation?
The model uses the SMIRK tokenizer, which requires Rust to be installed in the computational environment.
How are sequence embeddings pooled before the regression head?
The encoder hidden states are pooled by selecting the hidden state representation of the first token in the sequence.
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 preprint arXiv:2510.18900.
Written from MIST 1.8B · α (polarizability)'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.