Comparative Analysis of pK _a Predictions for Arsonic Acids Using Density Functional Theory-Based and Machine Learning...

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• Arsonic acids are environmental contaminants with variable ionization states affecting their properties and risks.
• Accurate prediction of arsonic acid pKa is essential but challenging due to limitations in existing models.

Purpose of the Study:

• To comparatively analyze the accuracy of pKa predictions for arsonic acids using machine learning and density functional theory models.
• To identify the most effective computational approach for predicting arsonic acid pKa values.

Main Methods:

• Support vector machine-based machine learning (ML) approach.
• Three density functional theory (DFT)-based models: V_S,max, SM D-based atomic charges, and scaled solvent-accessible surface.
• Comparative analysis of prediction accuracy using mean unsigned errors.

Main Results:

• The scaled solvent-accessible surface method showed high errors, indicating low effectiveness.
• The DFT method using atomic charges on the conjugated arsonate base provided the most accurate pKa predictions.
• The ML-based approach demonstrated strong predictive performance, suitable for broader chemical applications.
• The V_S,max DFT method showed weak prediction due to its limited scope in capturing molecular interactions.

Conclusions:

• The atomic charge-based DFT method and the ML approach are highly effective for predicting arsonic acid pKa.
• The V_S,max method is less reliable for pKa prediction as it oversimplifies molecular interactions.
• Accurate pKa prediction is vital for understanding and mitigating the environmental and health risks of arsonic acids.