Co-administration of drugs is a primary cause of Adverse Drug Reactions (ADRs) and a drain on the health care industry costing
billions of dollars and reducing quality of life. Drug-Drug Interactions (DDIs) account for as much as 30% of all ADRs. Unfortunately,
DDIs are not systematically explored pre-clinically and are difficult to detect in post-marketing drug surveillance. For this reason,
the detection and prediction of DDIs is an important problem in both drug development and pharmacovigilance. The comparison of the
3D drug structures provides a powerful tool for DDI prediction. In this article, we present the first large scale model for predicting DDIs
using the drug’s 3D molecular structure. In addition to identifying putative drug interactions we can also isolate the pharmacological or
clinical effect associated with the predicted interactions. The model has good performance in two different hold-out validations and in external
test sets. We found that the top scored drug pairs were significantly enriched for known clinically relevant interactions and that 3D
structure data is providing significantly independent information from other approaches, including 2D structure (p=0.003). We demonstrated
the usefulness of the proposed methodology to systematically identify pharmacokinetic and pharmacodynamic interactions, provided
an exploratory tool that can be used for patient safety and pre-clinical toxicity screening, and reviewed the state of the art methods
used to detect DDIs.
Keywords: 3D structure, adverse drug event, drug-drug interaction, pharmacophore, shape screening.
Rights & PermissionsPrintExport