Here, we provide the scripts used to model and analyze ionic liquid conductivity using the Nernst-Einstein hydrodynamic transport model, the modified Arrhenius kinetic transport model, t-stochastic neighbor embedding (t-SNE) dimensionality reduction, and machine learning modeling with various forms of chemical information inputs (molecular connectivity, 3D molecular descriptors, and bulk properties). We provide a framework for analyzing poorly-understood ionic liquid properties using easily accessible chemical information. In this work, we first contrast the accuracies of the Nernst-Einstein model and modified Arrhenius model predictions to understand which mechanistic model best describes ion transport in ionic liquids. We then use t-SNE to project ionic liquid molecular structure into a visualizable 2D latent space and analyze structure-conductivity relationships. Finally, we use machine learning models to test the our ability to predict ionic liquid conductivity using common chemical descriptors, and we analyze our models to identify which molecular descriptors and bulk properties most influence ion transport.

The databases we created for this analysis can be found in the Databases folder and combines RDKit descriptors for simulated single ions, reported PubChem molecular descriptors, and experimentally measured bulk properties from ILThermo. Multiple datasets are available for the various ionic liquid properties modeled; however our primary database of interest for conductivity contains 2,371 temperature dependent data points for 218 ionic liquids. Tutorials for our framework and results can be found in the Jupyter Notebooks folder.

Features

• Data science framework for analyzing ionic liquid properties
• Databases from RDKit, PubChem, and ILThermo containing properties and molecular descriptors for 218 ionic liquids and 2,371 temperature dependent data points
• Bulk property predictions for ionic liquids given their SMILES strings and RDKit Simulations
• Descriptor analysis to evaluate predictive capabilities and identify underlying ion transport mechanisms
• Classical model analyses to contrast ion transport mechanisms in ionic liquids

Publication

• Digital Discovery

Classical and Machine Learning Modeling

t-SNE Property Mapping

For detailed usage, navigate to the Jupyter Notebooks directory.

• PubChemPy v1.0.4
• RDKit v2024.3.5
• ILThermo v2.0
• pyILT2 v0.9.8

• Zavalab
• Gebbie Lab
• RDKit
• PubChem 3D
• IL Thermo