pKalculator: A pK _a predictor for C-H bonds

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Predicting C-H acidity (pKa) is crucial for synthetic chemists but becomes challenging with increasing molecular complexity.
• Accurate pKa values guide the prediction of reactive sites in organic synthesis.

Purpose of the Study:

• To develop an automated workflow (pKalculator) using quantum chemistry (QM) for calculating C-H pKa values.
• To generate a robust dataset for training a machine learning (ML) model to predict C-H pKa.
• To assess the model's performance in predicting pKa and identifying reaction sites.

Main Methods:

• A QM-based workflow, pKalculator, was developed for automated C-H pKa computations.
• The QM workflow was validated against 695 experimental C-H pKa values in DMSO.
• An ML model was trained on a dataset of 775 molecules containing 3910 C-H sites.

Main Results:

• The ML model achieved a mean absolute error (MAE) of 1.24 and a root mean squared error (RMSE) of 2.15 pKa units for C-H pKa predictions.
• The model successfully identified reaction sites in 1043 pKa-dependent reactions (aldol, Claisen, Michael) with a Matthew's correlation coefficient (MCC) of 0.82.

Conclusions:

• pKalculator provides an efficient and accurate method for determining C-H pKa values.
• The developed ML model demonstrates high predictive power for both C-H acidity and reaction site identification.
• This approach facilitates synthetic planning and reaction prediction in complex organic molecules.