Systems approaches to drug repositioning

Abstract

Drug discovery has overall become less fruitful and more costly, despite vastly increased biomedical knowledge and evolving approaches to Research and Development (R&D). One complementary approach to drug discovery is that of drug repositioning which focusses on identifying novel uses for existing drugs. By focussing on existing drugs that have already reached the market, drug repositioning has the potential to both reduce the timeframe and cost of getting a disease treatment to those that need it. Many marketed examples of repositioned drugs have been found via serendipitous or rational observations, highlighting the need for more systematic methodologies. Systems approaches have the potential to enable the development of novel methods to understand the action of therapeutic compounds, but require an integrative approach to biological data. Integrated networks can facilitate systems-level analyses by combining multiple sources of evidence to provide a rich description of drugs, their targets and their interactions. Classically, such networks can be mined manually where a skilled person can identify portions of the graph that are indicative of relationships between drugs and highlight possible repositioning opportunities. However, this approach is not scalable. Automated procedures are required to mine integrated networks systematically for these subgraphs and bring them to the attention of the user. The aim of this project was the development of novel computational methods to identify new therapeutic uses for existing drugs (with particular focus on active small molecules) using data integration. A framework for integrating disparate data relevant to drug repositioning, Drug Repositioning Network Integration Framework (DReNInF) was developed as part of this work. This framework includes a high-level ontology, Drug Repositioning Network Integration Ontology (DReNInO), to aid integration and subsequent mining; a suite of parsers; and a generic semantic graph integration platform. This framework enables the production of integrated networks maintaining strict semantics that are important in, but not exclusive to, drug repositioning. The DReNInF is then used to create Drug Repositioning Network Integration (DReNIn), a semantically-rich Resource Description Framework (RDF) dataset. A Web-based front end was developed, which includes a SPARQL Protocol and RDF Query Language (SPARQL) endpoint for querying this dataset. To automate the mining of drug repositioning datasets, a formal framework for the definition of semantic subgraphs was established and a method for Drug Repositioning Semantic Mining (DReSMin) was developed. DReSMin is an algorithm for mining semantically-rich networks for occurrences of a given semantic subgraph. This algorithm allows instances of complex semantic subgraphs that contain data about putative drug repositioning opportunities to be identified in a computationally tractable fashion, scaling close to linearly with network data. The ability of DReSMin to identify novel Drug-Target (D-T) associations was investigated. 9,643,061 putative D-T interactions were identified and ranked, with a strong correlation between highly scored associations and those supported by literature observed. The 20 top ranked associations were analysed in more detail with 14 found to be novel and six found to be supported by the literature. It was also shown that this approach better prioritises known D-T interactions, than other state-of-the-art methodologies. The ability of DReSMin to identify novel Drug-Disease (Dr-D) indications was also investigated. As target-based approaches are utilised heavily in the field of drug discovery, it is necessary to have a systematic method to rank Gene-Disease (G-D) associations. Although methods already exist to collect, integrate and score these associations, these scores are often not a reliable re flection of expert knowledge. Therefore, an integrated data-driven approach to drug repositioning was developed using a Bayesian statistics approach and applied to rank 309,885 G-D associations using existing knowledge. Ranked associations were then integrated with other biological data to produce a semantically-rich drug discovery network. Using this network it was shown that diseases of the central nervous system (CNS) provide an area of interest. The network was then systematically mined for semantic subgraphs that capture novel Dr-D relations. 275,934 Dr-D associations were identified and ranked, with those more likely to be side-effects filtered. Work presented here includes novel tools and algorithms to enable research within the field of drug repositioning. DReNIn, for example, includes data that previous comparable datasets relevant to drug repositioning have neglected, such as clinical trial data and drug indications. Furthermore, the dataset may be easily extended using DReNInF to include future data as and when it becomes available, such as G-D association directionality (i.e. is the mutation a loss-of-function or gain-of-function). Unlike other algorithms and approaches developed for drug repositioning, DReSMin can be used to infer any types of associations captured in the target semantic network. Moreover, the approaches presented here should be more generically applicable to other fields that require algorithms for the integration and mining of semantically rich networks.