Machine Learning for Drug Discovery: Predicting Drug-Protein Binding Affinities using Graph Convolutional Networks

Determining the potential interactions between ligands and proteins is a crucial first step in developing new drugs. Graph BAR, our novel prediction model, makes accurate guesses about the strength of ligand-protein bindings by means of a graph convolutional neural network. When compared to graph convolutional neural networks, standard convolutional neural network models are more resource and time intensive to manage. The interaction between proteins and ligands is shown via a network consisting of many adjacency meshes. The atomic distances dictate the alterations to these matrices. Additionally, the atomic properties can be shown in a feature matrix. Using the PDBbind datasets, we verified the functionality of graph convolution and evaluated Graph BAR's ability to forecast the binding strengths of proteins and ligands. In order to maximize The model's performance was improved by utilizing ways for adding additional input, which took advantage of the processing capability of graph convolutional neural networks. Graph BAR's predictions appear to have potential for use in drug creation due to its great accuracy and quickness. The results of our research suggest that Graph Convolutional Networks have the potential to expedite the discovery of new medications by properly predicting the interactions between pharmaceuticals and proteins.