Prediction of Drug-Drug Interactions Using Pharmacological Similarities of Drugs

Detection of potential Drug-Drug Interactions (DDIs) can reduce the costs associated drug administration and drug developments. It can also prevent serious adverse drug reactions possibly causing death. In this work, we have employed Rooted Page Rank algorithm in DDI network with weights calculated using therapeutic, genomic, phenotypic and chemical similarity of drugs to discover unknown DDIs. Weighting approach is inspired from the method used in collaborative filtering to score for recommendation of an item to a user based on similarities of users or items. Different than our previous work, this method enables the integration of global structure of DDI network with similarity scores of interactions to predict new DDIs. We obtained significant performance enhancement both in terms of AUC and Precision on DDI networks extracted from Drugbank. Interestingly some weighting scheme increases AUC and decreases precision such as in case of applying chemical similarity weighting. However, weighting with drug genomic similarities decreases AUC and raises precision. Therapeutic and phenotypic similarity weighting has increased performance of both in AUC and precision.