Drug-drug interaction networks are a great opportunity to use graph deep learning techniques to address the urgent healthcare problem of adverse drug interactions. In the United States alone, around 27.2% of adults have multiple chronic conditions for which they take multiple medications [1]. Chemical interactions between drugs can be quite dangerous; for example, mixing a sleep sedative such as doxylamine with an antihistamine such as diphenhydramine (commercial name Benadryl) can slow reaction times and make it risky to drive. Even if the drug interactions are not fatal, they can cause undesirable side effects or reactions which reduce quality of life. Therefore, it is vital that clinicians are aware of possible drug interactions before prescribing medications to patients.
To investigate drug-drug interactions, we will apply graph ML techniques to the ogbl-ddi dataset [2, 3], a homogeneous, unweighted, and undirected graph representing a drug-drug interaction network. This graph contains 4,267 nodes and 1,334,889 edges. Nodes represent FDA-approved or experimental drugs, and edges represent potential interactions between drugs. Notice that this dataset has no features beyond pure graph structure!