Multi-Modal Foundation · mist-models
MIST 1.8B · ZPVE
A 1.8-billion-parameter molecular transformer predicting zero-point vibrational energy from SMILES strings.
What MIST 1.8B · ZPVE is
MIST 1.8B · ZPVE is a molecular property prediction model developed by the Electrochemical Energy Group at the University of Michigan, Ann Arbor. It is part of the Molecular Insight SMILES Transformers (MIST) family of foundation models, which use masked language modelling over chemical structures to capture molecular features for downstream tasks.
This variant is fine-tuned to predict zero-point vibrational energy (ZPVE)—the quantum mechanical energy that remains in a molecule at absolute zero temperature. Accurately determining ZPVE is essential for thermochemical calculations, and predicting it directly from structural representations provides a fast alternative to resource-intensive quantum chemical simulations.
How it works
The model architecture consists of a RoBERTa-PreLayerNorm encoder with 24 layers, a hidden dimension of 2304, an intermediate size of 9216, 18 attention heads, and a maximum sequence length of 2048 tokens. For downstream property prediction, the encoder is paired with a task network consisting of a two-layer multilayer perceptron (MLP) with Gaussian Error Linear Unit (GELU) activations and dropout. The representation is formed by pooling the hidden state of the first token.
Pre-training was conducted on Kekulized SMILES strings from the Enamine REAL Space dataset using a masked language modelling objective with cross-entropy loss and the FusedLAMB optimizer across 32 NVIDIA A100 GPUs. The model was subsequently fine-tuned on the QM9 Zero Point Vibrational Energy dataset from the MoleculeNet benchmark on a single NVIDIA A100 GPU using mean squared error loss and the AdamW optimizer to output regression values in Hartree.
Key facts
- Base architecture
- RoBERTa-PreLayerNorm encoder with 2-layer MLP head
- Parameters
- 1.8 billion
- Pre-training dataset
- Enamine REAL Space
- Fine-tuning dataset
- QM9 Zero Point Vibrational Energy (MoleculeNet)
- Input format
- Kekulized SMILES strings
- Prediction target
- Zero-point vibrational energy (Hartree)
- Context length
- 2048 tokens
What you can use it for
- Zero-Point Vibrational Energy Estimation
- Predicting molecular ZPVE in Hartree directly from chemical structures to assist thermochemical calculations without running quantum chemical density functional theory simulations.
- Molecular Feature Extraction
- Extracting pooled vector representations from input SMILES sequences for downstream clustering, visualisation, or statistical learning tasks.
- High-Throughput Thermochemical Screening
- Evaluating candidate molecules in batches using string representations to quickly identify compounds with specific ground-state vibrational profiles.
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 · ZPVE
Strengths and limits
Good at
- Built on a large foundation encoder pre-trained over synthetic chemical space from Enamine REAL Space.
- Provides direct regression predictions of quantum thermochemical properties from simple text-based SMILES inputs.
- Accommodates molecular sequence lengths of up to 2048 tokens.
Watch out for
- Supervised fine-tuning is restricted to the QM9 benchmark, which comprises relatively small organic molecules.
- Inputs must be provided as Kekulized SMILES strings to match the representation seen during training.
- Requires 24 GB of GPU memory when deployed through standard feature-extraction runners.
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 · ZPVE 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 · ZPVE 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 used for the output predictions?
The model outputs zero-point vibrational energy predictions in Hartree.
What representation format must input molecules use?
Inputs must be provided as Kekulized SMILES strings, tokenised using the SMIRK tokenizer.
What is zero-point vibrational energy?
Zero-point vibrational energy (ZPVE) is the lowest possible quantum mechanical vibrational energy that a molecular system retains, even at absolute zero temperature, and is necessary for computing accurate thermochemical values.
What data was used to train this model?
The base encoder was pre-trained on molecular structures from Enamine REAL Space, after which the model was fine-tuned on the QM9 Zero Point Vibrational Energy benchmark.
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 · ZPVE'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.