Year: 2026
Type: preprint
Abstract: The protonation states of molecules are tightly linked to their physicochemical properties. Accurately predicting protonation states is therefore important for applications such as docking-based virtual screening in drug discovery. Modern machine-learning models can predict molecular p K a values with good accuracy, but obtaining pH-dependent protonation-state ensembles from available open-source models can cause substantial installation and workflow overhead. Commercial tools provide convenient alternatives but can entail substantial licensing costs. We here present pKasso, an accurate, fast, accessible, extensible, and easy-to-use open-source package to compute pH-dependent protonation-state ensembles of small molecules. pKasso enumerates relevant protonation microstates, integrates the existing p K a -prediction models MolGpKa and Uni-pKa, and converts their predictions into microstate populations and macroscopic p K a values. In addition, pKasso can combine predictions from multiple p K a -prediction models and report their disagreement as a qualitative indicator of predictive uncertainty. We evaluate pKasso on six benchmark datasets, where it achieves prediction accuracy competitive with state-of-the-art models. pKasso supports single-molecule and batch workflows. It is available as a web server ( tools.bindresearch.org/pkasso ), a Python library (pip install pkasso), and a command-line interface ( github.com/bindresearch/pkasso ).
Source: ChemRxiv
Institution: University College London
Language: English
Subfield: Drug Discovery
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7225835380