A Comparative Analysis of Artificial Intelligence Techniques for Modelling Drug Pharmacokinetics

Abstract

This work aims to predict the pharmacokinetic parameter which is Protein Binding (FB) through a comparative analytical study focusing on the relationship between drug and proteins, using the best 208 molecular descriptors of 263 active substances derived from RDKit as a sample. The study uses an artificial intelligence technique, namely machine learning techniques, including artificial neural networks (ANN) and a support vector machine coupled with the dragonfly algorithm (DA-SVM), trained on a reduced database of 25 molecular descriptors accessed through the genetic algorithm (GA). This database also uses reduced data exclusively for ANN and DA-SVM models. It has been found that there is an interactive relationship between the drug and proteins, especially plasma proteins such as albumin and alpha-1 acid glycoprotein, which affects the distribution, metabolism and elimination of the drug. Moreover, new deep learning methods have been discovered namely convolutional neural networks with multi-layer perceptron (CNN-MLP) and convolutional neural networks with DA-SVM (CNN-DASVM). The CNN-MLP model remarkably shows excellent and robust prediction ability, with criteria values of determination coefficient R2= 0.99 and RMSE= 0.0314. This study contributes to the advancement of pharmacokinetic modelling and highlights the potential of deep learning techniques in pharmaceutical research