Multi-Modal Foundation · mist-models
MIST 1.8B · Lipo
Predicts octanol-water distribution coefficients (logD) from molecular SMILES strings using a 1.8B transformer.
What MIST 1.8B · Lipo is
MIST 1.8B · Lipo is a molecular foundation model developed by the Electrochemical Energy Group at the University of Michigan. It specialises in predicting lipophilicity—specifically the octanol–water distribution coefficient (logD at pH 7.4)—directly from molecular structure strings.
Lipophilicity is a critical physicochemical property influencing how compounds permeate biological membranes, dissolve, and distribute throughout an organism. By adapting a large-scale chemical language model to experimental lipophilicity data, the model supports computational screening and property evaluation for small-molecule drug candidates and other functional compounds.
How it works
The model employs a RoBERTa-PreLayerNorm encoder architecture comprising 24 layers, an embedding dimension of 2,304, 18 attention heads, an intermediate feed-forward size of 9,216, and a context window of 2,048 tokens. During pre-training, it learned masked language representations over Kekulized SMILES strings drawn from the Enamine REAL Space database, optimised using FusedLAMB and cross-entropy loss.
For property prediction, the pre-trained encoder is paired with a task-specific network consisting of a two-layer multi-layer perceptron (MLP) with Gaussian Error Linear Unit (GELU) activations and dropout. The hidden state of the sequence's first token is pooled and passed through this MLP, which was fine-tuned on the MoleculeNet Lipophilicity dataset using Mean Squared Error loss and the AdamW optimizer.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder with a 2-layer MLP task head
- Parameters
- 1.8 billion
- Pre-training data
- Enamine REAL Space
- Fine-tuning data
- MoleculeNet Lipophilicity (4,200 small molecules)
- Input format
- Kekulized SMILES strings
- Output format
- Scalar logD at pH 7.4 or molecular embedding vector
- Context length
- 2,048 tokens
What you can use it for
- Lipophilicity Prediction
- Estimate octanol–water distribution coefficients (logD at pH 7.4) for candidate small molecules to assess solubility and membrane permeability.
- Virtual Screening
- Score large libraries of Kekulized SMILES representations to filter compounds with targeted lipophilicity profiles during lead discovery.
- Molecular Feature Extraction
- Extract pooled 2,304-dimensional sequence embeddings from the encoder to represent chemical structures for downstream analysis.
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 · Lipo
Strengths and limits
Good at
- Pre-trained on the synthetic compound library Enamine REAL Space, providing broad coverage of synthetically accessible chemical space.
- Handles long molecular string representations with an input context length of up to 2,048 tokens.
- Fine-tuned directly on measured logD distribution coefficients across 4,200 diverse small molecules.
Watch out for
- Supervised fine-tuning is limited to the chemical diversity present within the 4,200 molecules of the MoleculeNet Lipophilicity dataset.
- Expects Kekulized SMILES strings rather than arbitrary chemical string formats.
- Requires a Rust installation in the computational environment to compile and execute the SMIRK tokenizer.
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 · Lipo 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 · Lipo 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 chemical input does this model require?
The model takes molecular SMILES strings as input, specifically expecting them to be in Kekulized format.
What property value does the model output?
The regression head outputs a single scalar value predicting the octanol–water distribution coefficient (logD) measured at pH 7.4.
Can this model be used to generate molecular embeddings?
Yes, it can run as a feature-extraction model to return single pooled latent vectors representing the input chemical sequences.
What tokenizer does the model use?
It relies on the SMIRK tokenization package, which requires an active Rust compiler setup on the host system.
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 · Lipo'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.