Classification Prediction of ADMET Properties for Anti-Pancreatic Cancer Drug Candidates Using Machine Learning Approaches

Utilizing machine learning algorithms, this study aimed to predict the pharmacokinetic properties (ADMET properties) of anti-pancreatic cancer drug candidates, focusing on five key indicators: Absorption, Distribution, Metabolism, Excretion, and Toxicity. Based on 729 molecular descriptors derived from 1,974 compounds, classification prediction models were developed for intestinal epithelial cell permeability (Caco-2), cytochrome P450 3A4 subtype (CYP3A4) inhibition, cardiac safety (hERG), oral bioavailability (HOB), and micronucleus (MN) test. Various algorithms, including decision trees, discriminant analysis, support vector machines (SVMs), K-nearest neighbors, and ensemble learning, were applied. The Quadratic SVM algorithm emerged as the best performer in predicting Caco-2, CYP3A4, and hERG, achieving accuracies of 91.6%, 95.0%, and 90.5%, respectively. In contrast, the Boosted Tree model within ensemble learning showed superior performance in predicting HOB and MN, with accuracies of 90.5% and 96.1%. These findings demonstrate that machine learning algorithms significantly enhance the efficiency and success rate of drug development in the context of anti-pancreatic cancer candidates.