**3.3 Signature matching**

It is defined as the unique characteristics or "signature" of a drug which upon comparison with another drug, disease or clinical phenotype can yield another purpose of the drug [53]. Uniqueness of a signature owes to its chemical structure or changes transcriptomic, proteome, metabolome, or adverse event profiles that are generated upon its administration. Matching the signatures can be used to make drug-disease comparisons (estimating drug-disease similarity) [54] and drug-drug comparisons (drug-drug similarity) [55] and the correlation between the two defines the potential effect of drug on the disease [56]. Publicly accessible gene expression data of drugs and diseases have been mapped for easier drug repurposing predictions. Such an application is Connectivity Map (cMap), established in 2006 by the Broad Institute, and has been a success to predict drug-disease interactions. Other repositories such as Gene Expression Omnibus and Array Express that contain raw gene expression data from hundreds of disease conditions based on chemical structures with that of another drug to see whether there are chemical similarities could suggest shared biological activity. Upon selecting a set of chemical features for each drug a network is constructed based on the shared chemical features and is called the statistics-based cheminformatics. This approach was undertaken by Keiser and colleagues [2] to predict new targets for 878 FDA-approved small-molecule drugs and 2787 pharmaceutical compounds. Another such approach called similarity ensemble approach (SEA) evaluated the structural similarity of drug to target's ligand set, which led them to identify 23 new drug-target associations. But this approach has its limitations of errors in chemical structures and their physiological effects [54]. The signature-based approach has limitation of difficulty in mining adverse effect information from drug package inserts and the lack of well-defined adverse effect profiles and causality assessments for a number of drugs. However, artificial intelligence technologies that can undertake text mining and natural language processing represent potential future opportunities to overcome these limitations.
