excess-properties

What excess-properties is

excess-properties is a curated dataset compiled by Alexius Wadell and Anoushka Bhutani, designed to characterise non-ideal thermodynamic deviations in chemical mixtures. Mixture properties can be decomposed into a linear mixing baseline and an excess term that reflects molecular interactions between components. Quantifying these excess behaviours is essential for understanding mixture thermodynamics and identifying additives that modify physical properties even at very low concentrations.

The dataset contains 888,045 sparse observations across 715 unique molecules and 1,519 binary mixtures. Measurements include densities, molar enthalpies, molar volumes, and electrical conductivities, reported as functions of temperature and chemical composition.

The collection was created to fine-tune MIST mixture foundation models to predict thermodynamic deviations from ideal mixing behaviour, providing an empirical basis for chemical and materials science applications such as electrolyte solvent design.

How it was made

Data were manually transcribed from chemical literature and databases such as ILThermo using an open-source data entry interface. When original studies reported either mixture properties or excess properties alongside pure compound measurements, missing excess or mixture values were computed directly. The curators only utilised pure compound values reported within the specific source paper and avoided imputing values from external sources.

To streamline data collection, molar volumes were calculated from reported densities, relying on molecular weights derived using RDKit rather than transcribing reported molar volumes. Because original sources measured different physical properties, the compiled dataset is sparse across individual measurements.

Key facts

What you can use it for

Foundation model fine-tuning

Fine-tuning MIST chemical foundation models to predict non-ideal thermodynamic deviations and excess terms across varied chemical mixtures.

Electrolyte solvent formulation

Screening additive and solvent combinations for battery electrolytes by evaluating excess density and non-ideal intermolecular interactions.

Generalisation benchmarking

Assessing machine learning architectures on out-of-distribution molecules or unseen mixture pairs using compound-strict and mixture-strict cross-validation splits.

Explore the data

The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.

Strengths and limits

Questions

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, arXiv:2510.18900 [physics], 2025.