Electrostatic Interaction Force–Based Features Using PointNet Framework in Predicting RSK2 Inhibitor Bioactivity

Read full text Summarise with VIRAT
Authors: Tian Jiao Zhao, Rong-Ling Gu, Shijie Du, Lu Xu
Cited by: 0
References: 47

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Ribosomal S6 kinase 2 (RSK2) is a key mediator of cellular signal transduction. Given its well‐established roles in regulating cell proliferation, transformation, cell cycle progression, and apoptosis, RSK2 has become an important therapeutic target in oncology. To develop a broadly applicable quantitative structure‐activity relationship (QSAR) model for predicting RSK2 inhibitory potency, electron density for a series of inhibitors was computed using density functional theory (DFT). Using the cutting‐edge 3D point cloud framework PointNet, a novel molecular descriptor driven by quantum mechanics was introduced: PointNet features based on the Electrostatic Interaction Force (PNEIF) among molecular electron density points. Conventional descriptors were integrated with hybrid descriptors that incorporate PNEIF, and a particle swarm optimization–tuned least squares support vector machine (PSO‐LSSVM) model was built to predict the bioactivity of 724 potential RSK2 inhibitors; its performance was comprehensively evaluated. The hybrid model achieved 89.28% accuracy, markedly outperforming a traditional QSAR model (75.44%) and showing superior results across all evaluation metrics. Compared with conventional QSAR descriptors, PNEIF provides clear advantages by unifying molecular electron density with 3D structural information and, through a PointNet‐inspired architecture, capturing global features that are inherently invariant to rotation and translation. As a novel 3D descriptor, PNEIF is well suited for modeling complex biological targets, delivering improved performance on challenging samples, improved cross‐structural generalization, and greater interpretability of the selected features.

Fields of Study & Subjects

Computational Drug Discovery Methods Advanced Chemical Physics Studies Machine Learning in Materials Science

Publication details

Journal / source
Journal of Chemistry
Volume
2026
Issue
1
Pages
4899563
Publication date
1 January 2026
Language
EN