Screened Electrolyte Solvents for Battery Safety from Molecular Structure

Over one billion molecules were screened to identify safer nonaqueous battery solvents using graph neural networks. The dataset, created by Stephen R. Xie and last updated in April 2026, contains predictions for flash point, melting point, dielectric constant, and liquid viscosity. Candidates were sourced from the Mcule catalog of existing molecules and the GDB-13 enumeration of hypothetical molecules.

Use Cases

• Training machine learning models to predict solvent properties based on molecular structure.
• High-throughput screening for novel, safer battery electrolyte candidates.
• Identifying readily available molecules from commercial catalogs for experimental validation.
• Exploring the chemical space of hypothetical molecules for de novo electrolyte design.

Strengths

• Screens a massive chemical space of over one billion molecules.
• Predicts four key solvent properties (flash point, melting point, dielectric constant, viscosity) directly from structure.
• Uses a two-pronged approach covering both existing and hypothetical molecules for diverse candidates.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the source databases (DIPPR 801, Mcule, GDB-13).

Provenance

Source

figshare

Collection Method

Predictions from graph neural networks trained on the DIPPR 801 database, applied to molecules from the Mcule catalog and GDB-13.

Freshness

Last updated 2026-04-20 10:14:38