Multi-Modal Foundation · mist-models
MIST 27.1M · ETN
A molecular transformer predicting the normalized Dimroth-Reichardt solvent polarity parameter from SMILES strings.
What MIST 27.1M · ETN is
MIST 27.1M · ETN 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, designed for exploration and property prediction across chemical space.
The model addresses the task of estimating solvent polarity, specifically predicting the normalized Dimroth-Reichardt solvent polarity parameter (ETN). This empirical polarity parameter is widely used in chemistry to describe solvent effects, solvent-solute interactions, and chemical reactivity in liquid media.
Built by fine-tuning the pretrained MIST-28M foundation encoder (specifically mist-28M-ti624ev1), the model processes chemical structures provided as text strings and outputs continuous regression values for polarity, enabling rapid property assessment for prospective solvents without direct laboratory measurement.
How it works
The base architecture utilizes a RoBERTa-PreLayerNorm encoder consisting of 8 layers, 8 attention heads, a hidden size of 512, an intermediate size of 2048, and a maximum sequence context length of 2048 tokens. For pretraining, the model was trained on SMILES strings from the Enamine REAL Space database using a Masked Language Modeling (MLM) objective with cross-entropy loss, optimised using FusedLAMB.
For property prediction, the encoder is paired with a task network composed of a two-layer multi-layer perceptron (MLP) featuring Gaussian Error Linear Units (GELU) activations and dropout. Input molecules are supplied as Kekulized SMILES strings. Consistent with standard transformer pooling, the task network takes the hidden state of the first sequence token and is fine-tuned for regression via Mean Squared Error (MSE) loss using the AdamW optimiser on normalized ETN values sourced from the Stenutz solvent polarity table.
Key facts
- Base architecture
- RoBERTa-PreLayerNorm encoder (8 layers, 512 hidden size, 8 attention heads) with a 2-layer MLP head
- Context length
- 2048 tokens
- Pretraining dataset
- Enamine REAL Space
- Fine-tuning dataset
- Stenutz solvent polarity table (normalized Dimroth-Reichardt ETN values)
- Primary input
- Kekulized SMILES strings
- Primary output
- Normalized Dimroth-Reichardt solvent polarity parameter (ETN)
What you can use it for
- Solvent polarity screening
- Predicting normalized Dimroth-Reichardt ETN values directly from molecular SMILES to screen candidate solvents and solvent-like media.
- Chemical formulation design
- Evaluating the empirical polarity characteristics of proposed organic molecules for electrolyte, formulation, or reaction medium design.
- Molecular representation extraction
- Extracting pooled token embedding vectors or mean-pooled representations from the trained encoder for downstream machine learning tasks.
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 27.1M · ETN
Strengths and limits
Good at
- Pretrained on the Enamine REAL Space dataset, exposing the encoder to a vast collection of synthetically accessible molecules.
- Accepts standard textual SMILES strings, allowing rapid in silico inference without requiring three-dimensional molecular conformers.
- Fine-tuned directly on curated empirical solvent polarity data from the Stenutz polarity database.
Watch out for
- Inputs are expected to be formatted as Kekulized SMILES strings; alternative representations may cause degraded behaviour.
- Evaluation is performed on an 80/10/10 random split of the Stenutz dataset, which may share chemical scaffolds across splits.
- Requires the custom SMIRK tokenizer, which relies on a Rust installation environment to operate.
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 27.1M · ETN 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 27.1M · ETN 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 property does this model predict?
It predicts the normalized Dimroth-Reichardt solvent polarity parameter (ETN), a standard empirical index of solvent polarity and solvatochromism.
What is the expected input format for chemical structures?
The model expects molecular structures supplied as Kekulized SMILES strings.
What base model was used to create this checkpoint?
It was created by fine-tuning mist-28M-ti624ev1, an 8-layer RoBERTa-PreLayerNorm encoder pretrained on Enamine REAL Space.
How are the sequence embeddings pooled before prediction?
The encoder hidden states are pooled by taking the hidden state vector of the first token, which is then fed to the two-layer MLP prediction head.
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 [physics], 2025. https://doi.org/10.48550/arXiv.2510.18900
Written from MIST 27.1M · ETN'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.