Multi-Modal Foundation · mist-models
MIST 1.8B · ESOL
A molecular foundation model fine-tuned to predict aqueous solubility from chemical SMILES strings.
What MIST 1.8B · ESOL is
MIST 1.8B · ESOL is a molecular regression model developed by the Electrochemical Energy Group at the University of Michigan, Ann Arbor. It predicts the aqueous solubility of small molecules directly from their chemical structures represented as SMILES strings.
Aqueous solubility is an essential physicochemical parameter that influences chemical processing, formulation, and biological availability. Experimental solubility measurements can be labour-intensive and resource-heavy, creating a need for automated estimation directly from molecular sequences.
The model builds on the 1.8-billion-parameter Molecular Insight SMILES Transformer (MIST) pre-trained on large-scale chemical structures from Enamine REAL Space. It adapts the encoder with a dedicated prediction head to deliver single-target regression estimates of aqueous solubility.
How it works
The model uses a RoBERTa-PreLayerNorm encoder architecture consisting of 24 layers, a hidden dimension of 1536, an intermediate size of 6144, and 24 attention heads, supporting sequences up to 2048 tokens. It was pre-trained using a masked language modelling objective on SMILES strings derived from the Enamine REAL Space database, constructed through forward synthetic analysis from experimentally validated building blocks.
For downstream solubility prediction, the pre-trained encoder is coupled to a task network consisting of a two-layer multi-layer perceptron (MLP) with Gaussian Error Linear Units (GELU) activations and dropout. Hidden states are pooled using the representation of the first token, and the network is fine-tuned using Mean Squared Error loss and the AdamW optimizer on the ESOL dataset from the MoleculeNet benchmark.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder with a 2-layer MLP regression head
- Parameter count
- 1.8 billion (encoder)
- Encoder layers
- 24
- Attention heads
- 24
- Hidden size
- 1536
- Maximum sequence length
- 2048
- Pre-training data
- Enamine REAL Space
- Fine-tuning data
- ESOL dataset from MoleculeNet (1,128 small molecules)
What you can use it for
- Aqueous Solubility Prediction
- Estimate the experimental aqueous solubility (log S in mols per litre) for small organic molecules from their Kekulized SMILES strings.
- High-Throughput Candidate Screening
- Screen batches of chemical candidates, such as drug compounds, fragrances, or electrolytes, for suitable water solubility before synthesis or testing.
- Molecular Feature Extraction
- Extract high-dimensional mean-pooled or sequence-level representations of chemical SMILES strings for downstream computational chemistry workflows.
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 · ESOL
Strengths and limits
Good at
- Leverages a massive 1.8-billion-parameter foundation model pre-trained on diverse, synthetically accessible chemical space from Enamine REAL Space.
- Accepts sequence lengths up to 2048 tokens, accommodating complex and extended SMILES strings.
- Uses pre-layer normalization and standard transformer structures optimised for stable representations.
Watch out for
- Fine-tuning was conducted on the ESOL dataset containing only 1,128 small molecules, which may limit generalisation across novel or highly complex chemical scaffolds.
- Requires Kekulized SMILES inputs processed via the SMIRK tokenizer, which relies on a Rust installation.
- Evaluated specifically on single-target aqueous solubility regression, meaning it cannot predict other molecular or thermodynamic properties without additional fine-tuning.
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 · ESOL 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 · ESOL 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 format should input molecules use?
Inputs must be provided as molecular SMILES strings, specifically Kekulized SMILES strings as used in pre-training and fine-tuning.
What is the exact output of the model?
The model outputs a regression prediction corresponding to measured log aqueous solubility (log S) in mols per litre, or pooled representation vectors when run in feature-extraction pipelines.
Which benchmark was used to evaluate this model?
The model was fine-tuned and tested on the ESOL (Estimated SOLubility) regression dataset from MoleculeNet using an 80/10/10 random data split and Root Mean Squared Error (RMSE) as the evaluation metric.
What tokeniser is required to run the model?
The model uses SmirkTokenizerFast from the smirk package, which requires a working Rust environment to install and run.
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 1.8B · ESOL'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.