چكيده به لاتين
Abstract:
Today, with the advances made in the field of biology, it is possible to model
biological data. The same has led to the emergence of a variety of biological
networks such as protein-protein interaction network, drug-protein-target network,
drug-drug interaction network, and so on. Because of the large volume of data and
the complexity of them and their complex structure, it is not possible to manually
analyze these networks. For this reason, the use of graph mining solutions to analyze
this type of data has been considered. The main objective of this study is to
investigate the various biological networks in order to extract the hidden relationship
among drug side effects. In this study, since there are no database as the side effectside effect-interaction, we have challenge for evaluating our proposed method in
computer science domain. Thus, first, we implement traditional link prediction
techniques like simrank, jaccard and so on, to predict relationships among side effect
as baseline algorithms. Second, we combine link prediction techniques and
clustering method as the proposed method to discover hidden relationships among
drug side effects. Third, we compare proposed method and link prediction methods
as a baseline and indicate, the proposed method predict some relationships that
traditional link prediction methods cannot predict them. We show cluster level
analysis that consider relationships among drugs as feature of drug side effects is
better than node level link prediction that consider only drugs as feature of drug side
effects, in extracting relationships among side effects. Finally, for validating results
in biological domain, we search top score extracted relationships in biological
articles from PubMed.
Keywords: Graph mining, Clustering, Link prediction, Bioinformatics, Drug side
effects, Adverse drug reaction
