H-SPOC pKa Prediction

Molecular SMILES will be shown here.

#	Solvent	H_Index	pK_a

Usage

1. Draw the target molecule using Ketcher (left) and then click "Generate 3D Structure".

2. Wait for preprocessing until 3D structure appears in 3DMol.js (right). It generally takes 10-30 second, may last longer for large molecules.

3. Select target solvents and the target site (H index) using the pull-down menu, respectively; and then click "Start Prediction".

4. The prediction may take 5-30 seconds.

Model Introduction: H-SPOC is an XGBoost machine-learning model for site-specific pK_a prediction in molecular and supramolecular systems across 39 different solvents. It shows the state-of-art performance in SAMPL6, SAMPL7 and SAMPL8 challenges as of Feb, 2025.

Dataset: H-SPOC is based on iBonD experimental database, (http://ibond.nankai.edu.cn.) including 18,812 experimental data for 14,838 small molecules. The dataset covers 39 different solvent types and 27 types of functional groups with pK_a in a range of 12.5 to 40.0 units.

Notice: For extremely weak acids such as simple arene or alkane C-H acids, please use the prediction with cautions or contact us for interpretation before any usage.

Cite this :Liu, S., Yang, Q., Zhang, L., Luo, S. Highly Precise Prediction of Micro- and Supra-pK_a Based on 3D Descriptors Integrating Non-Covalent Interactions. Angew. Chem. Int. Ed. 2025, e202424069.