A systematic pathway-based network approach for in silico drug repositioning

Drug repositioning, the method of finding new uses for existing drugs, holds the potential to reduce the cost and time of drug development.
Successful drug repositioning strategies depend heavily on the availability and aggregation of different drug and disease databases.
Moreover, to yield greater understanding of drug prioritisation approaches, it is necessary to objectively assess (benchmark) and compare different methods.

Data aggregation requires extensive curation of non-standardised drug nomenclature.
To overcome this, we used a graph-theoretic approach to construct a drug synonym resource that collected drug identifiers from a range of publicly available sources, establishing missing links between databases.
Thus, we could systematically assess the performance of available in silico drug repositioning methodologies with increased power for scoring true positive drug-disease pairs.

We developed a novel pathway-based drug repositioning pipeline, based on a bipartite network of pathway- and drug-gene set correlations that captured functional relationships.
To prioritise drugs, we used our bipartite network and the differentially expressed pathways in a given disease that formed a disease signature.
We then took the cumulative network correlation between disease pathway and drug signatures to generate a drug prioritisation score.
We prioritised drugs for three case studies: juvenile idiopathic arthritis, Alzheimer's and Parkinson's disease.
We explored the use of different true positive lists in the evaluation of drug repositioning performance, providing insight into the most appropriate benchmark designs.

We have identified several promising drug candidates and showed that our method successfully prioritises disease-modifying treatments over drugs offering symptomatic relief.
We have compared the pipeline’s performance to an alternative well-established method and showed that our method has increased sensitivity to current treatment trends.
The successful translation of drug candidates identified in this thesis has the potential to speed up the drug-discovery pipeline and thus more rapidly and efficiently deliver disease-modifying treatments to patients.