Machine learning methods for drug repurposing

Abstract

Drug Repurposing consists on using already approved drugs to treat other diseases. This is done by identifying new targets that the drug may have. To accelerate this long and costly process, computational methods have been developed to predict drug-target interactions (DTIs). Recently, Machine Learning has had a tremendous impact on many scientific fields including DTI prediction. In this project, two state-of-the-art methods named MolTrans and Hyper- AttentionDTI are described and compared on 8 different datasets. Moreover, each dataset is divided according to 3 different splits, so that the generalization of the methods with respect to drugs and proteins is tested. Graphs are a type of data structures that have nodes and edges connecting them. They can model complex systems accurately such as DTI networks. In recent years, Graph Machine Learning methods have been developed to improve on conventional models. Node embedding techniques and Graph Neural Networks are introduced as ways to transform nodes into vector embeddings to be able to make predictions on them. While the two analyzed methods offer competitive results, Graph Machine Learning can take advantage of the expressiveness that graphs have in order to make more accurate predictions.