Multi-Modal Foundation · mist-models
MIST 1.8B · LUMO
A 1.8-billion-parameter model predicting molecular LUMO energy from SMILES strings for chemical property discovery.
What MIST 1.8B · LUMO is
MIST 1.8B · LUMO is a molecular property prediction model developed by the Electrochemical Energy Group at the University of Michigan, Ann Arbor. It fine-tunes the 1.8-billion-parameter Molecular Insight SMILES Transformer (MIST) foundation model to predict the lowest unoccupied molecular orbital (LUMO) energy of molecules directly from textual chemical representations.
LUMO energy is a quantum chemical property indicating the energy level of a molecule's lowest unoccupied molecular orbital, which governs its electron affinity and chemical reactivity. By mapping molecular structures represented as Simplified Molecular Input Line Entry System (SMILES) strings to LUMO values in Hartree, the model allows researchers to rapidly query molecular electronic characteristics without running demanding quantum mechanical simulations.
How it works
The model is built on a RoBERTa-PreLayerNorm encoder featuring 24 layers, 24 attention heads, a hidden size of 1536, and an intermediate size of 6144, supporting sequences up to 2048 tokens. It was initially pre-trained on Kekulized SMILES strings from the Enamine REAL Space compound database using a masked language modelling (MLM) objective and cross-entropy loss, trained with the FusedLAMB optimiser across 32 NVIDIA A100 GPUs.
For property prediction, the pre-trained encoder is coupled to a task network consisting of a two-layer multilayer perceptron with Gaussian Error Linear Unit (GELU) activations and dropout. The hidden state corresponding to the first token of the sequence is pooled into a single representation vector and passed to the regression head. The fine-tuning phase was trained on the QM9 dataset using mean squared error (MSE) loss and the AdamW optimiser on a single NVIDIA A100 GPU.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder with a two-layer MLP task network
- Parameter count
- 1.8 billion
- Context length
- 2048 tokens
- Pre-training data
- Enamine REAL Space
- Fine-tuning data
- QM9 (LUMO energy)
- Prediction target
- LUMO energy (Hartree)
What you can use it for
- LUMO Energy Estimation
- Predicting the lowest unoccupied molecular orbital energy in Hartree directly from molecular SMILES notation.
- Reactivity and Electron Affinity Screening
- Evaluating candidate molecules in chemical libraries based on predicted electron-accepting capability and reactivity trends.
- Molecular Representation Extraction
- Generating dense embedding vectors from the encoder's pooled token representations for general chemical informatics tasks.
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 · LUMO
Strengths and limits
Good at
- Pre-trained on the Enamine REAL Space database, exposing the base model to synthetically accessible chemical structures.
- Predicts quantitative quantum chemical properties in Hartree directly from 1D SMILES strings without needing 3D conformer generation.
- Accommodates molecular sequence lengths of up to 2048 tokens.
Watch out for
- Fine-tuned on the QM9 dataset, which limits the supervised calibration primarily to small organic molecules.
- Expects molecular inputs to be formatted specifically as Kekulized SMILES strings.
- Requires the Smirk tokenizer library, which depends on 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 1.8B · LUMO 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 · LUMO 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 input format does this model require?
The model takes molecular SMILES strings as input, specifically expecting Kekulized SMILES strings processed through the SmirkTokenizerFast library.
In what unit are the predicted LUMO values expressed?
Predictions are continuous scalar values expressed in Hartree.
How are token embeddings pooled before regression?
The encoder hidden states are pooled by taking the hidden state vector of the first token in the sequence, which is then fed into the two-layer MLP task network.
How was the model evaluated during development?
It was evaluated using an 80/10/10 random split of the QM9 dataset, measuring predictive performance via Mean Absolute Error (MAE).
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 · LUMO'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.