pKalculator: A pK_a predictor for C–H bonds

• Rasmus M. Borup,
• Nicolai Ree and
• Jan H. Jensen

Beilstein J. Org. Chem. 2024, 20, 1614–1622, doi:10.3762/bjoc.20.144

• . Furthermore, we employ our model on 1043 pKa-dependent reactions (aldol, Claisen, and Michael) and successfully indicate the reaction sites with a Matthew’s correlation coefficient (MCC) of 0.82. Keywords: C–H pKa values; pKa predictor; Introduction Over the years, the ability to selectively break a C–H

• , for example, steric strain and charge delocalisation. We discuss this further in Supporting Information File 1, section “Outliers for the test set”. We then compare our ML model with previously reported ML models for predicting pKa values, namely, the GCNN C–H pKa predictor by Roszak et al. [3] and

• the XGBoost pKa predictor by Yang et al. [5]. Roszak et al. [3] used a mix of experimental data (414 compounds) [7], manually curated DFT data (212 compounds), and previously reported DFT data (194 C–H sites) [11]; they obtained a MAE of 2.18 pKa units for their test set. Yang et al. [5] used filtered