Expanding the chemical space of ionic liquids using conditional variational autoencoders

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Ionic liquids (ILs) are versatile salts with low melting points, offering tunable solvent properties for catalysis, batteries, and drug delivery.
• The vast chemical space of ILs remains largely unexplored, limiting the development of advanced applications.
• Current machine learning models for IL generation are constrained by limited ion diversity in existing databases.

Purpose of the Study:

• To develop a machine learning approach for generating novel and diverse ionic liquids (ILs).
• To address the limited ion diversity in existing IL databases for improved generative model performance.
• To expand the explored chemical space of ILs with a focus on low-melting-point compounds.

Main Methods:

• Utilized conditional variational autoencoders (CVAEs) for generating novel cations and anions.
• Introduced a novel ion scoring method to prioritize ions likely to form low-melting-point ILs.
• Developed a melting point prediction model to identify promising cation-anion pairs.

Main Results:

• Successfully generated diverse and novel cations and anions using CVAEs.
• Identified cation-anion pairs likely to form ILs with low melting points.
• Validated through molecular dynamics simulations that 13 out of 15 generated ILs have desirable low melting points (<373 K).

Conclusions:

• The developed CVAE approach effectively expands the chemical space of ILs with novel structures.
• The ion scoring method enhances the generation of ILs with targeted properties, specifically low melting points.
• The generated ILs show significant potential for various applications, validated by predictive models and simulations.