UpKast: Method, Model, and Database for Large-Scale Cross-Solvent Determination of pKa Values
Abstract
The acid dissociation constant, commonly written as p K a , plays a key role in chemistry. However, most p K a measurements have been conducted in only a handful of solvents, resulting in data scarcity for many relevant solvent systems. Herein, we introduce the UpKast-DB, which comprises of p K a estimates for around 3,000 acids and bases in 29 solvents. We constructed UpKast-DB using a quantum-chemical method that integrates solvation free energy calculations with experimental data. Furthermore, we trained a surrogate machine learning model UpKast-ML on the computed values, allowing for rapid and accurate prediction of p K a values beyond the acids and bases presented in UpKast-DB. The quantum-chemical and machine learning methods result in RMSEs of approximately 1 p K a unit relative to experimental values, slightly higher than the typical aleatoric uncertainty of underlying experimental data, but more accurate than many existing experimental measurements (which can have uncertainties up to, and sometimes beyond, 3 p K a units). The data and modeling approaches provide a means of productively augmenting existing p K a data sources, and represent a significant step toward general determination of p K a in any solvent. The dataset and models can be found on Zenodo: doi:10.5281/zenodo.23239570
Related
- Structural, electronic, spectroscopic, and thermodynamic properties of gentisic acid: solvent and temperature effects using DFT and TD-DFT approaches
- Transport of organic solutes in ion-exchange membranes: Mechanisms and influence of solvent ionic composition
- ConSolv: Solvent-Conditional Machine Learning Implicit Solvent Potential
- Solvent-induced memory effects in a model electrolyte
- Size and solvent effects on cellulose transverse anisotropy and toughening design
BibTeX
@misc{zheng2026upkast,
title = {{UpKast: Method, Model, and Database for Large-Scale Cross-Solvent Determination of pKa Values}},
author = {Jonathan Wenyuan Zheng and William H. Green},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010254/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010254/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - UpKast: Method, Model, and Database for Large-Scale Cross-Solvent Determination of pKa Values
AU - Jonathan Wenyuan Zheng
AU - William H. Green
PY - 2026
DA - 2026/10/12
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010254/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010254/v1
AB - The acid dissociation constant, commonly written as p K a , plays a key role in chemistry. However, most p K a measurements have been conducted in only a handful of solvents, resulting in data scarcity for many relevant solvent systems. Herein, we introduce the UpKast-DB, which comprises of p K a estimates for around 3,000 acids and bases in 29 solvents. We constructed UpKast-DB using a quantum-chemical method that integrates solvation free energy calculations with experimental data. Furthermore, we trained a surrogate machine learning model UpKast-ML on the computed values, allowing for rapid and accurate prediction of p K a values beyond the acids and bases presented in UpKast-DB. The quantum-chemical and machine learning methods result in RMSEs of approximately 1 p K a unit relative to experimental values, slightly higher than the typical aleatoric uncertainty of underlying experimental data, but more accurate than many existing experimental measurements (which can have uncertainties up to, and sometimes beyond, 3 p K a units). The data and modeling approaches provide a means of productively augmenting existing p K a data sources, and represent a significant step toward general determination of p K a in any solvent. The dataset and models can be found on Zenodo: doi:10.5281/zenodo.23239570
ER -