Multi-Modal Foundation · mist-models
MIST 28M · Lipo
Predicts the octanol-water distribution coefficient (logD at pH 7.4) of small molecules from SMILES strings.
What MIST 28M · Lipo is
MIST 28M · Lipo is a molecular property prediction model that estimates lipophilicity, quantified as the octanol-water distribution coefficient (logD at pH 7.4). Built by the Electrochemical Energy Group at the University of Michigan, Ann Arbor, the model combines a pre-trained molecular transformer encoder with a task-specific regression head.
Lipophilicity is a critical physicochemical property that governs how chemical compounds partition between aqueous and organic environments, affecting the pharmacokinetics of therapeutic candidates and the behaviour of electrolyte formulations. This model enables high-throughput estimation of logD directly from string-based molecular representations.
How it works
The underlying architecture is a RoBERTa-PreLayerNorm encoder with 8 transformer layers, 8 attention heads, a hidden size of 512, an intermediate size of 2048, and a maximum context length of 2048 tokens. The encoder was pre-trained on Kekulized SMILES from the Enamine REAL Space dataset using a masked language modelling objective, cross-entropy loss, and the FusedLAMB optimizer across two NVIDIA A100-SXM4-80GB GPUs.
For downstream lipophilicity prediction, the pre-trained encoder is coupled with a two-layer multi-layer perceptron featuring Gaussian Error Linear Unit (GELU) activations and dropout. The model pools the sequence by extracting the hidden state of the first token. The full architecture was fine-tuned on an NVIDIA A100 GPU using mean squared error loss and the AdamW optimizer on the MoleculeNet Lipophilicity dataset, which contains experimental measurements for 4,200 compounds.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder with a two-layer MLP task head
- Parameter count
- 28M
- Pre-training dataset
- Enamine REAL Space
- Fine-tuning dataset
- MoleculeNet Lipophilicity (4,200 molecules)
- Maximum sequence length
- 2048 tokens
- Input format
- Kekulized SMILES
What you can use it for
- Octanol-water distribution prediction
- Estimates logD values at pH 7.4 for small molecules to assess how they partition across aqueous and lipid phases.
- Molecular representation extraction
- Outputs pooled vector representations from the first token hidden state for downstream cheminformatics workflows.
- Candidate molecule screening
- Evaluates lipophilic properties of compounds within synthetically accessible chemical spaces for drug discovery and electrolyte research.
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 · Lipo
Strengths and limits
Good at
- Accommodates long molecular inputs up to a context limit of 2048 tokens.
- Pre-trained on the extensive, synthetically accessible Enamine REAL Space chemical database.
- Fine-tuned directly on experimentally validated logD distribution measurements.
Watch out for
- Molecules generally must be formatted as Kekulized SMILES strings prior to evaluation.
- Fine-tuning performance was evaluated using an 80/10/10 random split rather than a scaffold-based split.
- Tokenisation relies on the SMIRK tokenizer, which requires a local Rust installation.
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 · 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 28M · 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 CPU only, typically in 12s, and a run costs about 28 credits. Visitors without an account get 1,000 credits a month.
What molecular input format does the model require?
The model takes SMILES strings as inputs, which generally need to be Kekulized before being passed to the tokenizer.
What prediction does the regression head generate?
The model predicts the octanol-water distribution coefficient (logD at pH 7.4) measured in logD units.
Which tokeniser is required to run the model?
The model uses the SMIRK tokenizer (SmirkTokenizerFast), which requires Rust to be installed on the system.
How was the fine-tuning performance assessed?
The model was evaluated using a random 80/10/10 train, validation, and test split on the MoleculeNet Lipophilicity dataset, scored by Root Mean Squared Error (RMSE).
How to cite it
Wadell, Alexius, Anoushka Bhutani, Victor Azumah, Austin R. Ellis-Mohr, Celia Kelly, Hancheng Zhao, Anuj K. Nayak, Kareem Hegazy, Alexander Brace, Hongyi Lin, Murali Emani, Venkatram Vishwanath, Kevin Gering, Melisa Alkan, Tom Gibbs, Jack Wells, Lav R. Varshney, Bharath Ramsundar, Karthik Duraisamy, Michael W. Mahoney, Arvind Ramanathan, and Venkatasubramanian Viswanathan. 'Foundation Models for Discovery and Exploration in Chemical Space.' arXiv:2510.18900 (2025).
Written from MIST 28M · 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.