Multi-Modal Foundation · mist-models
MIST 28M · solvent properties
A 28-million-parameter transformer predicting physical and solvatochromic descriptors of organic solvents from SMILES.
What MIST 28M · solvent properties is
MIST 28M · solvent properties is a molecular property prediction model developed by the Electrochemical Energy Group at the University of Michigan, Ann Arbor. Built upon the MIST (Molecular Insight SMILES Transformers) foundation model family, it targets bulk physical and chemical descriptors of organic solvents from chemical structure strings.
The model addresses the experimental overhead of characterising solvent environments by predicting key solvatochromic and polarity scales. These scales, such as Catalan and Kamlet-Taft parameters, quantify how solvents interact through hydrogen bonding, polarisability, and dipolarity, which are critical for reaction engineering, formulation, and electrolyte design.
How it works
The model employs a RoBERTa-PreLayerNorm architecture comprising 8 layers, 8 attention heads, a hidden size of 512, and an intermediate size of 2048, capped at a maximum sequence length of 2048 tokens. The underlying encoder was pre-trained on Kekulised Simplified Molecular Input Line Entry System (SMILES) strings from the Enamine REAL Space dataset using a masked language modelling objective with cross-entropy loss.
For property estimation, the pre-trained encoder is paired with a task network consisting of a two-layer multi-layer perceptron (MLP) using Gaussian Error Linear Unit (GELU) activations and dropout. Downstream regression fine-tuning relies on mean squared error loss, taking the pooled hidden state of the first sequence token to generate continuous solvent property predictions.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder with a 2-layer MLP head
- Parameter count
- 28M
- Pre-training data
- Enamine REAL Space
- Maximum context length
- 2048 tokens
- Input format
- Kekulised SMILES strings
What you can use it for
- Solvent polarity estimation
- Predicting empirical solvent polarity parameters, such as the normalised Dimroth-Reichardt parameter (ETN), directly from solvent molecular structures.
- Hydrogen bonding and polarisability profiling
- Inferring Kamlet-Taft solvatochromic parameters (alpha for hydrogen bond acidity and beta for hydrogen bond basicity) using chemical SMILES strings.
- Catalan parameter prediction
- Predicting Catalan solvatochromic scales, including SPP1 (solvent polarity/polarisability), SB3, and SA4, to assess bulk physical behaviour.
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 28M · solvent properties
Strengths and limits
Good at
- Pre-trained on Enamine REAL Space, allowing the encoder to leverage synthetically accessible chemical space representations.
- Uses lightweight 28-million-parameter representations, enabling efficient inference on standard single-GPU hardware.
- Accepts standard Kekulised SMILES strings directly as inputs for rapid regression predictions.
Watch out for
- Fine-tuning on Kamlet-Taft parameters relied on a small dataset of only 182 values measured by NMR.
- The internal pKa prediction head is deprecated in favour of a dedicated DMSO pKa model.
- Requires the SMIRK tokenizer, which depends on a Rust installation to run.
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 28M · solvent properties 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 28M · solvent properties 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 input format does this model expect?
The model expects molecular structures provided as Kekulised SMILES strings tokenised using the SMIRK tokenizer.
Which solvent scales does the model cover?
It is fine-tuned to predict the normalised Dimroth-Reichardt polarity parameter (ETN), Catalan solvatochromic parameters (SPP1, SB3, SA4), and Kamlet-Taft parameters (alpha and beta).
Should this model be used for pKa predictions?
No. The pKa outputs on this checkpoint are deprecated, and the separate mist-models/mist-28M-6zlgl2qn-pKa model should be used instead for pKa in DMSO.
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 preprint arXiv:2510.18900 (2025). https://doi.org/10.48550/arXiv.2510.18900
Written from MIST 28M · solvent properties'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.