Title: ADMET property prediction model based on feature selection and data mining techniques

Authors: Junlin Gu; Yihan Xu; Juan Sun; Weiwei Liu

Addresses: College of Computer, Jiangsu Vocational College of Electronics and Information, Huai'an, 223003, China ' College of Computer, Jiangsu Vocational College of Electronics and Information, Huai'an, 223003, China ' College of Computer, Jiangsu Vocational College of Electronics and Information, Huai'an, 223003, China ' College of Computer, Jiangsu Vocational College of Electronics and Information, Huai'an, 223003, China

Abstract: Breast cancer has posed a significant threat to women's health in recent years, and the search for compounds that can antagonise ERα activity will play an important role in breast cancer treatment. While previous studies employed machine learning to predict ADMET properties, performance constraints persisted. In this paper, we present a data mining method to establish a predictive model for biologically active ADMET properties. We first use feature selection methods to identify 23 impactful variables. After that, the LightGBM and genetic algorithms are applied to perform biological activity prediction. Furthermore, the ADMET-UMLP model, utilising a BP neural network with a U-shaped structure, effectively leveraged feature information. The model performed well on the validation set, with AUC values exceeding 0.9 in the classification prediction of Caco-2, CYP3A4, hERG, HOB, and MN properties, and a prediction of 0.98 AUC value for Caco-2, demonstrating good predictive performance.

Keywords: ADMET; absorption; distribution; metabolism; excretion; toxicity; LightGBM; machine learning; prediction.

DOI: 10.1504/IJAACS.2025.148531

International Journal of Autonomous and Adaptive Communications Systems, 2025 Vol.18 No.4, pp.310 - 323

Received: 11 Jul 2023
Accepted: 12 Dec 2023
Published online: 11 Sep 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article