pKa prediction from an ab initio bond length: part 2--phenols

Area of Science:

• Computational Chemistry
• Physical Organic Chemistry
• Quantitative Structure-Property Relationships (QSPR)

Background:

• Accurate prediction of acidity constants (pK(a)) is crucial in chemistry.
• Existing methods often involve complex descriptors and models.
• The pursuit of simpler, yet accurate, predictive models remains an active research area.

Purpose of the Study:

• To investigate the potential of a single descriptor, specifically ab initio bond length, for predicting pK(a) values.
• To demonstrate that simpler models can yield more accurate predictions.
• To develop accurate predictive models for phenols with deviations less than 0.50 log units from experimental values.

Main Methods:

• Utilized a dataset of 171 phenols.
• Focused on the carbon-oxygen bond length connecting the hydroxyl group to the phenyl ring as the primary descriptor.
• Developed single-bond-length models, including the creation of high-correlation subsets for ortho-substituted phenols.
• Employed cross-validation to assess model performance.

Main Results:

• A single descriptor, the ab initio carbon-oxygen bond length, accurately predicts pK(a) for meta- and para-substituted phenols within 0.50 log units.
• Splitting ortho-substituted phenols into high-correlation subsets improved prediction accuracy.
• Single-bond-length models outperformed models using multiple descriptors and exhibited better cross-validation statistics (RMSEE < 0.50, RMSEP < 0.50).
• Outliers were readily identified and often explained.

Conclusions:

• A single ab initio bond length is a powerful and simple descriptor for predicting phenol pK(a) values.
• The use of high-correlation subsets enhances prediction accuracy for specific compound classes, like ortho-substituted phenols.
• This approach offers a robust and interpretable method for pK(a) prediction, facilitating outlier identification and model refinement.