Architecture of the drug-drug interaction network

Summary Purpose: Drug interaction information has been extensively compiled into large databases. The objective of the present study was to provide a systematic overview of the available drug interaction information, using a network approach. Methods: The drug–drug interaction information was retrieved from a comprehensive source reference that documents primary drug interaction information over an extended period of time. With careful examination of the information, we identified three continuously growing databases that consisted of 351, 636 and 966 drugs and 742, 1858 and 3351 pairs of interaction, respectively. We then constructed three drug–drug interaction networks in which the interacting drugs were treated as nodes and were connected with links that represent interactions. For each network, we determined the number of interactions that each drug in that network has, and prepared histograms to show the frequency distribution. Results: The frequency distribution or the probability that a given drug has k interactions, P(k), followed a power‐law distribution, where the power law exponent was close to ‐1·5 and was independent of the network size. The results suggested that while the majority of the drugs in the network had few interactions (small k), highly interacting drugs (large k) were rare but contributed most of the network interactions. Conclusions: The present study demonstrated that drug interaction information can be viewed and analysed as a connecting, growing network. As with many real‐world networks, the drug interaction network was scale free, indicating that drug interaction information has been dominated by a relatively small number of highly interacting drugs.