Multi-Modal Foundation · mist-models
MIST 1.8B · HOMO-LUMO gap
A 1.8-billion-parameter molecular transformer predicting the HOMO-LUMO gap from Kekulized SMILES strings.
What MIST 1.8B · HOMO-LUMO gap is
MIST 1.8B · HOMO-LUMO gap is a molecular property prediction model developed by the Electrochemical Energy Group at the University of Michigan, Ann Arbor. It pairs a 1.8-billion-parameter foundation encoder from the Molecular Insight SMILES Transformers (MIST) family with a downstream regression network specialised for quantum chemical property estimation.
The model specifically predicts the highest occupied molecular orbital to lowest unoccupied molecular orbital (HOMO-LUMO) gap of small molecules. This electronic energy difference is a fundamental quantum chemical parameter used to evaluate chemical reactivity, stability, and optical excitation characteristics.
The base encoder was pre-trained on large-scale chemical structures from the Enamine REAL Space dataset and subsequently fine-tuned on the QM9 benchmark from MoleculeNet to deliver regression predictions in units of Hartree directly from molecular SMILES notation.
How it works
The architecture consists of a RoBERTa-PreLayerNorm encoder with 24 layers, a hidden dimension of 1536, an intermediate size of 6144, 24 attention heads, and a maximum context sequence length of 2048 tokens. For pre-training, the encoder was trained on Kekulized SMILES strings using a masked language modelling (MLM) objective with cross-entropy loss, optimised via FusedLAMB across 32 NVIDIA A100 GPUs.
For property prediction, the pre-trained encoder is coupled to a task network comprising a two-layer multi-layer perceptron (MLP) with Gaussian Error Linear Unit (GELU) activations and dropout. Hidden states are pooled using the representation of the sequence's first token. Fine-tuning was carried out on QM9 HOMO-LUMO gap targets using Mean Squared Error (MSE) loss and the AdamW optimiser on a single NVIDIA A100 GPU.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder with a 2-layer MLP head
- Parameter count
- 1.8 billion
- Pre-training data
- Enamine REAL Space
- Fine-tuning data
- QM9 HOMO-LUMO gap dataset (MoleculeNet)
- Input format
- Kekulized SMILES strings
- Output format
- HOMO-LUMO gap in Hartree (regression) or mean-pooled feature vectors
- Maximum sequence length
- 2048 tokens
What you can use it for
- HOMO-LUMO Gap Prediction
- Estimate the orbital energy difference of organic molecules in Hartree directly from their Kekulized SMILES representations.
- Chemical Reactivity and Stability Screening
- Screen compound libraries for chemical reactivity and electronic stability based on predicted orbital energy separations.
- Molecular Representation Extraction
- Extract high-dimensional vector representations or sequence embeddings of small molecules for downstream cheminformatics pipelines.
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 · HOMO-LUMO gap
Strengths and limits
Good at
- Leverages a 1.8-billion-parameter representation pre-trained across extensive synthetic chemical space from Enamine REAL Space.
- Supports input sequences up to 2048 tokens, accommodating large molecular SMILES representations.
- Replaces costly density functional theory (DFT) simulations with rapid regression inference for HOMO-LUMO gap estimates.
Watch out for
- Tokenizer dependencies include the SMIRK tokenizer, which requires a Rust environment to build and run.
- Expects input molecules formatted strictly as Kekulized SMILES strings.
- Model weights are released for research purposes only without guarantees of correctness or fitness for purpose.
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 · HOMO-LUMO gap 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 · HOMO-LUMO gap 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 units are the HOMO-LUMO gap predictions reported in?
Outputs from the regression head are reported in Hartree.
What inputs does the model accept?
The model takes molecular SMILES strings, specifically requiring Kekulized SMILES representations.
How were the datasets split during fine-tuning?
The QM9 fine-tuning dataset was evaluated using a random 80/10/10 split for training, validation, and testing.
What software environment is needed for tokenisation?
The model relies on the SmirkTokenizerFast class from the smirk library, which requires the Rust toolchain to be installed.
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:2510.18900 [physics], 2025. https://doi.org/10.48550/arXiv.2510.18900
Written from MIST 1.8B · HOMO-LUMO gap'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.