Some Mathematical Perspective of Graph Neural Network

Abstract

Graph Neural Networks (GNNs) have become highly effective tools for molecular property prediction, which is essential for drug research, material science, and environmental science applications. The Researchers use GNN to increase the precision of molecular property predictions through the representation of molecules as graphs, in which bonds are edges and atoms are nodes, for GNNs are able to capture the intricate interactions seen in molecular structures. The Combining between deep learning and cheminformatics improves Quantitative Structure-Activity Relationships (QSAR) and other prediction models. The goal of this work is to present a novel approach to molecular property prediction with GNNs, which will overcome the drawbacks of conventional computational methods and lessen the necessity for lengthy experimental testing.