Dydruma: A Dynamic Drug Map for Exploring the Associations Among Drug Therapeutic Indications and Side-Effects

Inferring potential therapeutic indications and identifying clinically important side-effects are both important and challenging tasks in modern drug development. Previous studies have utilized drug chemical structures and protein targets to construct predictive models for both tasks. According to our study, the drug therapeutic information itself is highly predictive for side-effects, and drug side-effect information highly predictive of therapeutic indication. This confirms that there exist underlying associations among drug therapeutic indications and side-effects. Exploring these associations can lead to better understanding of the drugs as well as more informed hypotheses for drug repositioning and adverse effect monitoring. In practice however, it is impractical to check all possible associations using an exhaustive list. In this study, we present Dydruma, a dynamic drug map which encodes drug therapeutic indications and side-effects as well as their associations. The map can be dynamically adjusted based on a significance value attached to each association, where the significance value is derived from a statistical test such as Fisher’s exact test. We describe an optimization-based approach for dynamic bipartite graph layout to ensure visual continuity among successive layouts at different significance thresholds, to help the user maintain a consistent mental map throughout the exploration. We demonstrate the effectiveness of dytruma for exploring the associations among drug indications and side effects and for hypothesis generation using real world data sets.