Link Prediction in Bipartite Graphs and Its Applications to Drug-Target Interaction Prediction

The prediction of interactions between drugs (medicines) and pharmacological targets (proteins) is one of the most prominent applications of machine learning it the pharmaceutical industry. From the theoretical point of view, the task is to predict unknown links in a bipartite graph. Compared with standard link prediction, in case of drug-target interaction prediction, additional information is available, such as the similarity between pairs of drugs and the similarity between pairs of targets. The incorporation of this additional information is one of the key components of successful drug-target interaction prediction techniques. This talk will review some of the most prominent link prediction techniques and their adapta- tion to drug-target interaction prediction ranging from simple weighted profile (a.k.a. nearest neighbor or collaborative filtering) approaches over methods based on matrix factorisation to bipartie local models and some of its recent variants.