Ligand-Based Virtual Screening Using Graph Neural Network and Learning Representation

Abstract

Information technology has become an essential component of the predict molecular properties process and virtual screening (VS) has emerged as a valuable computational technique for efficiently and cost-effectively screening chemical compounds. A crucial task in VS is the similarity search, which aims to estimate the similarity between molecules. This approach is based on the assumption that molecules with similar structures may exhibit similar biological activities. Several techniques have been developed to compare the biological similarity between a target compound and each compound in a database. While these methods perform well, particularly for molecules with homogeneous active structures, they often struggle when dealing with structurally diverse compounds. Ligandbased virtual screening (LBVS) is a critical computational approach in drug development that aids in the identification of potential lead compounds with high affinity and selectivity for a target protein. Traditional LBVS methods often rely on molecular descriptors and similarity measures, which may not adequately capture the complex relationships between chemical structures and biological activities. In recent years, graph neural networks (GNNs) have emerged as a powerful tool for learning representations of molecular graphs and predicting their properties. This paper presents an abstract on the application of GNNs in LBVS for drug development, highlighting their potential to overcome the limitations of conventional methods and enhance the efficiency and effectiveness of lead compound identification