QupKake: Integrating Machine Learning and Quantum Chemistry for Micro-pK_a Predictions

Area of Science:

• Computational Chemistry
• Physical Chemistry
• Chemical Informatics

Background:

• Accurate prediction of micro-pKa values is essential for understanding molecular properties.
• Applications span drug discovery, materials science, and environmental chemistry.
• Existing models face challenges in accuracy and generalization.

Purpose of the Study:

• To introduce QupKake, a novel computational method for micro-pKa prediction.
• To combine graph neural network (GNN) models with semiempirical quantum mechanical (QM) features.
• To achieve high accuracy and generalization in micro-pKa prediction.

Main Methods:

• Development of the QupKake method integrating GNNs and QM features.
• Training and validation on diverse benchmark datasets.
• Analysis of feature importance to understand model contributions.

Main Results:

• QupKake demonstrated superior performance compared to state-of-the-art models.
• Achieved root-mean-square errors between 0.5 and 0.8 pKa units on external test sets.
• QM features were identified as critical for reaction site enumeration and prediction.

Conclusions:

• QupKake offers a significant advancement in micro-pKa prediction accuracy and generalization.
• The method provides a powerful tool for chemical research and development.
• Highlights the importance of integrating QM features in predictive models.