Predicting pK_a Using a Combination of Semi-Empirical Quantum Mechanics and Radial Basis Function Methods.

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Machine Learning

Background:

• The acid dissociation constant (pKa) significantly impacts molecular properties essential for drug discovery and development.
• Accurate pKa prediction is crucial for optimizing compound synthesis, formulation, and pharmacokinetic properties (ADME).

Purpose of the Study:

• To develop and validate a novel computational method for accurately predicting pKa values.
• To combine semi-empirical quantum mechanical calculations with machine learning for enhanced predictive power.

Main Methods:

• Utilized semi-empirical quantum mechanical calculations.
• Integrated machine learning algorithms to build a predictive model.
• Validated the model on diverse external datasets, including the SAMPL6 challenge and known drug compounds.

Main Results:

• Achieved excellent accuracy in pKa prediction, with root-mean-square errors between 0.7-1.0 log units.
• Demonstrated comparable performance to higher-level computational methods but with significantly reduced computational cost.
• Successfully predicted pKa for both mono- and polyprotic compounds.

Conclusions:

• The developed hybrid quantum mechanics/machine learning approach provides an efficient and accurate method for pKa prediction.
• This method holds promise for accelerating compound development in pharmaceutical research.
• Offers a cost-effective alternative to computationally intensive methods for pKa determination.