Network Clustering of Drug-Drug Interactions

Drug drug interactions (DDIs) are instances when one drug blunts the effectiveness of another and may lead to unpredictable side effects that threaten patient safety. Elucidating DDIs is paramount to the safety and effectiveness of the drug treatments. In this project, we applied network clustering to identify groups of drugs that tend to interact with one another, and determine potential biochemical similarities underpinning the interactions. Using network clustering - an unsupervised AI/machine learning method involving grouping of drugs based on their interactions within a network, we constructed a DDI network of more than 1 million known DDIs, with nodes depicting the drugs, and the edges depicting evidence of the DDIs. Using the Louvain Community algorithm, we clustered drugs into groups with more frequent interactions. To optimize clustering, inter- and intra-cluster analyses were performed to validate the structure. Categories of the drugs were mapped within these clusters to identify properties influencing DDIs. It was found that some significant determinants of DDIs include the drug structures, the part of the body or type of drug they target. These insights highlight the value of predicting DDIs in preventing adverse side effects. These properties underpin the DDI clusters and are valuable in the predictions of DDIs in the prevention of side effects.