Multi-Modal Foundation · mist-models
MIST 26.9M · odour
A molecular transformer predicting 135 scent descriptors from chemical SMILES strings.
What MIST 26.9M · odour is
MIST 26.9M · odour is a molecular property prediction model designed to predict olfactory qualities of chemical compounds. Developed by the Electrochemical Energy Group at the University of Michigan, Ann Arbor, the model is built on the Molecular Insight SMILES Transformer (MIST) architecture and fine-tuned specifically for odour profiling.
Determining how a molecule smells is an important challenge in fragrance formulation, flavour chemistry, and chemical exploration. Because an individual chemical structure can exhibit multiple sensory qualities at once, odour prediction is treated as a multi-label classification problem across many scent categories.
The model accepts Kekulized SMILES strings—textual representations of chemical structures—and outputs predictions across 135 scent labels. It provides computational chemists and sensory scientists with a quantitative method for assessing scent characteristics directly from molecular notation.
How it works
The underlying architecture consists of an 8-layer RoBERTa-PreLayerNorm encoder with a hidden dimension of 512, an intermediate size of 2048, 8 attention heads, and a maximum sequence length of 2048 tokens. For the odour prediction task, a two-layer multi-layer perceptron (MLP) task head with Gaussian Error Linear Units (GELU) activations and dropout receives the pooled hidden state from the first sequence token.
The encoder was pre-trained on molecular SMILES strings from the Enamine REAL Space database using a masked language modelling objective with cross-entropy loss and the FusedLAMB optimizer. It was subsequently fine-tuned on an olfaction dataset of 4,983 molecules compiled by Lee and colleagues, using binary cross-entropy loss and the AdamW optimizer for multi-label classification.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder with a two-layer MLP head
- Parameter count
- 26.9 million
- Context length
- 2048 tokens
- Pre-training dataset
- Enamine REAL Space
- Fine-tuning dataset
- Olfaction dataset (4,983 molecules)
- Input format
- Kekulized SMILES strings
- Output
- Logits for 135 scent descriptors
What you can use it for
- Olfactory profiling of small molecules
- Predict the presence or absence of 135 distinct scent descriptors for chemical candidates directly from their SMILES strings.
- Virtual screening for flavour and fragrance design
- Screen candidate compounds to identify molecules likely to express specific aroma notes such as floral, fruity, or woody characteristics.
- Molecular representation extraction
- Generate pooled sequence embeddings of chemical structures for downstream analysis and chemical space exploration.
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 26.9M · odour
Strengths and limits
Good at
- Supports multi-label classification across 135 sensory descriptors simultaneously.
- Pre-trained on Enamine REAL Space, a large database of commercially accessible chemical structures.
- Processes long chemical sequences up to a context length of 2048 tokens.
Watch out for
- Requires Kekulized SMILES strings as inputs rather than standard or canonical SMILES.
- Fine-tuning was conducted on a relatively small dataset of 4,983 molecules.
- Produces unnormalised logits that require calibration or thresholding to derive binary predictions.
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 26.9M · odour 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 26.9M · odour 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 does the model expect?
The model expects molecular SMILES strings, specifically Kekulized SMILES, formatted up to a maximum context length of 2048 tokens.
What outputs does the model generate?
It outputs unnormalised logits representing binary classification predictions across 135 scent descriptors.
Which tokenizer is required to prepare inputs?
The model uses the SMIRK tokenizer, which requires a Rust environment to be installed.
How was the model evaluated?
Evaluation was performed on a random 10% test split of the olfaction dataset, measuring the area under the receiver operating characteristic curve (AUROC) averaged across all 135 labels.
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].
Written from MIST 26.9M · odour'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.