Multi-Modal Foundation · mist-models
MIST 27.0M · conductivity
A molecular transformer that predicts ionic conductivity and thermodynamic parameters for liquid electrolyte mixtures.
What MIST 27.0M · conductivity is
MIST 27.0M · conductivity is a specialised molecular property prediction model designed to estimate the ionic conductivity of complex liquid electrolytes. Developed by the Electrochemical Energy Group at the University of Michigan, the model addresses the challenge of rapidly screening multi-component battery electrolyte formulations across varying temperatures and compositions without relying solely on computationally intensive physics-based simulations or physical laboratory measurements.
The model builds upon the MIST (Molecular Insight SMILES Transformers) foundation model family, taking chemical representations in Simplified Molecular Input Line Entry System (SMILES) format alongside mixture proportions and operating temperatures. It predicts the natural logarithm of ionic conductivity along with related thermodynamic parameters, assisting researchers in evaluating transport properties within single-salt ternary-solvent systems.
By capturing both composition-weighted molecular representations and temperature-dependent relations, the model enables high-throughput computational evaluation of liquid electrolyte formulations for energy storage and electrochemical applications.
How it works
The model employs a RoBERTa-PreLayerNorm encoder architecture consisting of 8 layers, 8 attention heads, a hidden size of 512, an intermediate size of 2048, and a maximum sequence length of 2048. The base encoder was pre-trained using a Masked Language Modelling (MLM) objective and cross-entropy loss on molecular SMILES strings from the Enamine REAL Space database, processed via SMIRK tokenisation.
For the fine-tuning stage, a two-layer multi-layer perceptron (MLP) task network is attached to the encoder. The task network forms a mixture embedding by summing individual component embeddings weighted by their respective mole ratios. This mixture vector predicts two sets of coefficients: one parameterising the Vogel-Fulcher-Tammann (VFT) relation to capture temperature dependence, and another parameterising an empirical correction term for concentration dependence. Fine-tuning used multi-channel regression optimised with AdamW and mean squared error loss, trained on 24,822 single-salt ternary-solvent electrolyte mixtures generated by the Advanced Electrolyte Model (AEM).
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder with a 2-layer MLP head
- Encoder layers
- 8
- Attention heads
- 8
- Hidden size
- 512
- Intermediate size
- 2048
- Maximum sequence length
- 2048 tokens
- Pre-training dataset
- Enamine REAL Space
- Fine-tuning dataset
- 24,822 single-salt ternary-solvent mixtures from Zhu et al. (Advanced Electrolyte Model)
- Pre-training objective
- Masked Language Modelling (MLM)
- Pre-training hardware
- 2 NVIDIA A100-SXM4-80GB GPUs (12 hours 15 minutes)
- Fine-tuning hardware
- 1 NVIDIA A100 GPU
What you can use it for
- Electrolyte Ionic Conductivity Screening
- Predict the natural logarithm of ionic conductivity (in ln(mS/cm)) for formulated battery electrolytes consisting of single salts dissolved in up to three solvents.
- Temperature Dependence Parameter Estimation
- Estimate the pseudo-activation energy (Ea) and Vogel temperature (Tg) in Kelvin to evaluate how electrolyte transport characteristics vary with temperature.
- Electrolyte Solvent Mixture Optimisation
- Vary solvent mole fractions and chemical structures across candidate formulations to computationally assess optimal composition ratios prior to synthesis.
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 27.0M · conductivity
Strengths and limits
Good at
- Incorporates domain-informed Vogel-Fulcher-Tammann (VFT) relationship parameterisation to capture physical temperature dependencies.
- Combines pre-trained structural representations with explicit component mole fractions via a weighted sum mixture embedding.
- Outputs physical transport parameters—pseudo-activation energy and Vogel temperature—alongside ionic conductivity.
Watch out for
- Trained specifically on single-salt mixtures containing up to three solvents, limiting applicability to more complex multi-salt or quaternary solvent systems.
- Requires tokenisation using SMIRK tokenizers, which requires a Rust installation in the execution environment.
- Fine-tuning was conducted on synthetic data generated by the Advanced Electrolyte Model (AEM) rather than purely experimental measurements.
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 27.0M · conductivity 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 27.0M · conductivity 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 inputs does the model require for inference?
The model expects a structured dictionary containing SMILES strings and mole fractions for up to three solvents, SMILES for the cation and anion of the salt, and the measurement temperature specified in Kelvin.
What specific property values are returned by the prediction method?
The model returns three values: the natural logarithm of ionic conductivity (ln(mS/cm)), the pseudo-activation energy Ea (in Kelvin), and the Vogel temperature Tg (in Kelvin).
What was the pre-training data source for the underlying encoder?
The encoder was pre-trained on the Enamine REAL Space database, a collection of commercially accessible chemical structures built using forward synthetic analysis.
How does the model represent chemical mixtures?
It computes individual molecular embeddings for the mixture components using the RoBERTa encoder and calculates a composite mixture embedding as the sum of component vectors weighted by their mole fractions.
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, 2025, arXiv:2510.18900 [physics]. https://doi.org/10.48550/arXiv.2510.18900
Written from MIST 27.0M · conductivity'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.