Deep learning has revolutionized the field of machine learning and artificial intelligence in the last few years, especially for structured or grid-format data, as well as for linear (Euclidean space) data. Graph Neural Networks (GNNs) are a promising area of research that has emerged in various domains such as physics, chemistry, biology, computer networks, social networks, traffic networks, recommender systems, and natural language processing. Such researches often involve graph-based representations of the molecules, atoms, and point-based structures, representing non-Euclidean space data. GNNs gain immense popularity due to their capability to minimize laboratory efforts, primarily in atom- and molecule-level simulations. GNN is a model of neural network representation of graphs, based on capturing node dependence and important relative parameters. It thereby helps make the precise prediction of node and edge attributes. This paper provides deeper insight into the recent research works on GNN-based research and studies future directions related to this relatively new field of study.