Evidence-based drug repurposing using graph neural networks

Abstract

Graph Neural Network is the class of Deep Learning algorithms that can work directly on Graphs and take advantage of their structural information. In real life, we all know that there are many such datasets that can be easily and efficiently represented in a network so we can take advantage of them using GCN here we are mainly focusing on using Graph Convolution Network (GCN) and Jumping Knowledge Network (JK-Net). Here we see that the dataset of ours contains genes & drugs is a network of how genes are affected by drugs, how the drugs interact among them, and how genes interact among themselves can be very easily represented by using Hetero-graph and DGL Graph Library so that any new relations can be easily learned from them using an algorithm that takes advantage of that structural information in those Graphs. Moreover what is the basis of this work is to merge those different Graphs into a single Heterogeneous-Graph and use the GCN algorithm to predict new relations between them. The goal of this work is to take advantage of the Graph Learning algorithm to predict links that will eventually lead to the Re-purposing of the Drugs which is comparatively an easier process than developing a new drug for a new variant of the disease which is a long process and will take not less than 10-15 years and a lot of resources and wealth but if the existing drugs can be repurposed successfully, will not only save time and effort but also lives.