Prediction of pKa Values for Neutral and Basic Drugs based on Hybrid Artificial Intelligence Methods

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• The pKa value of drugs is critical for drug design and pharmacological activity.
• Existing methods for predicting pKa values may lack accuracy or efficiency.
• Developing robust quantitative structure-activity relationship (QSAR) models is essential for drug discovery.

Purpose of the Study:

• To propose an improved particle swarm optimization (PSO) algorithm for enhanced QSAR model training.
• To develop a predictive model for drug pKa values using a radial basis function artificial neural network (RBF ANN).
• To evaluate the prediction performance and applicability of the developed model.

Main Methods:

• An improved PSO algorithm was developed, incorporating population entropy diversity with adaptive strategies (convergence, divergence, self-adaptive adjustment).
• The improved PSO algorithm was used to train a radial basis function artificial neural network (RBF ANN) model.
• Molecular descriptors were selected, and a QSAR model was established to predict pKa values for neutral and basic drugs.

Main Results:

• The developed RBF ANN model, trained with the improved PSO algorithm, demonstrated good prediction performance for drug pKa values.
• Validation using an independent dataset yielded an absolute average relative error of 0.3105, root mean square error of 0.0411, and squared correlation coefficient of 0.9685.
• The model accurately predicted pKa values for 74 neutral and basic drugs.

Conclusions:

• The improved PSO algorithm effectively enhances RBF ANN training for QSAR modeling.
• The developed QSAR model provides accurate predictions of drug pKa values, serving as a valuable tool for drug design.
• This approach can be referenced for exploring other QSAR relationships in pharmacology.