Acidity Prediction in Arbitrary Solvents: Machine Learning Based on Semiempirical Molecular Orbital Calculation.

Area of Science:

• Computational Chemistry
• Physical Chemistry
• Machine Learning Applications

Background:

• Solvent effects nonlinearly impact acid behavior, necessitating precise solvent selection for applications.
• A significant gap exists in predicting acid dissociation constants (pKa) across varied molecular structures and solvents.
• Current limitations hinder the prediction of acidity in arbitrary chemical environments.

Purpose of the Study:

• To develop a versatile protocol for predicting pKa in diverse environments.
• To integrate quantum chemical calculations with machine learning for accurate acidity prediction.
• To address the challenge of nonlinear solvent effects on acid strength.

Main Methods:

• Utilized quantum chemical calculations with a polarizable continuum model.
• Developed machine learning models trained on a limited dataset.
• Applied the protocol to predict pKa for biologically relevant acids and superstrong acids in organic solvents.

Main Results:

• Achieved an average absolute error of 1.1 for pKa predictions in both aqueous and organic solvents.
• Successfully modeled the nonlinear "compression effect" of acidity decay with solvation.
• Demonstrated the protocol's accuracy for molecules with complex electronic changes upon proton dissociation.

Conclusions:

• The developed protocol offers accurate and versatile pKa prediction across a wide range of compounds and environments.
• This approach overcomes limitations in predicting acidity, supporting fields like drug discovery and chemical engineering.
• The method effectively accounts for complex solvation effects on acid strength.