Objective

Background One successful method of reducing time, cost and failure rate for introducing new treatments to patients is to give existing drugs new indications, known as drug repurposing. This is mostly achieved via serendipity, but a systematic approach can be taken by pharmacologists, biologists and clinicians by matching opposing genomic signatures.

Objectives To pragmatically review free to use Genomic Based Drug Repurposing websites.
Methods A standardised pro-forma was used to compare each website. An experimentally validated mouse epilepsy signature was used to obtain drug lists. The drug results were validated and compared using data on all prescribed drugs that have evidence of working in Epilepsy models, but not approved for this indication. Exclusion: charges for use, disease specific, requires software download, requires non-genomic data.
Results Eight websites met the eligibility criteria. L1000CDS2 and CLUE are the authors’ recommendations for first time-users. L1000CDS2 provides the necessary information for interpreting the results, it was the only website to link to clinical trial evidence. However, validation of the results using this paper’s methods is not possible. CLUE has significantly larger pools of data to use, however interpreting the results is not straight-forward. Data Source: CLUE contained the highest number of drug signatures to search (250479), GDA the highest number of unique compounds to search (50816) and NFFinder the highest number of cell lines to search (410). Four websites used GEO as a data source, three used LINCS and interestingly LINCS accounted for >80% of the drug signatures used overall. Methodology of search: Gene input in standard HGNC format was used by six websites, seven websites used lists of up- and down- regulated genes. Five websites provided significance scores for the results. Six websites gave the experimental details for each result. Usability: Search results were mostly returned within seconds, the interpretation of results is time-consuming and lacks evidence-based recommendations. Overall, websites have room for significant usability improvements for non-technical and non-computational users. A worrying number of websites are not well maintained, 3 of 11 websites became inactive during the four-month study period. Validation and comparison of drug lists: 6 of 8 websites did not provide sufficient data or sufficiently organised data for validation and comparison. CLUE and CMAP02 gave no significant enrichment of results when correcting for multiple testing. Only two websites stated which drug naming format was used, six appeared to use multiple rare annotations. A crude comparison was only possible for the other websites. The proportion of search results that are prescribe-able and have evidence of working in Epilepsy: CLUE, 5/18; CMAP02, 1/4; GDA, 1/1; NFFinder, 1/5; PDOD, 2/10; L1000CDS2, 0/3; L1000FWD, 2/6.

Conclusions Access to every website's key data and reasonable interoperability of such data is urgent. Drug names in tools should be standardised to make comparison more effective. Due to these factors no effective validation could be performed. The authors’ opinion is that clinical staff and biologists should be involved further in website development to ensure they are usable and useful. Engagement and usage of GBR websites by clinical staff and biologists will be vital for realising their patient benefit.