disease gene expression signatures were compared with drug-induced gene expres-

sion profiles obtained from the connectivity map (Lamb et al. 2006) to derive a

therapeutic score. Drugs with significant negative scores have gene expression

patterns that are anti-correlated with disease-specific gene expression patterns and

therefore represent putative novel therapeutic indications (Dudley et al. 2011).

4.3.2

Side Effect Similarity

Drug side effects are the result of complex phenotypes that arise due to a number of

molecular interactions including the interaction with the primary target or off-targets

(Campillos et al. 2008). Although off-target interactions of the existing drugs are

generally undesired and harmful, they can occasionally be useful and can lead to

development of new therapeutic options for drugs (e.g., sildenafil). The drugs

lacking chemical similarity can cause similar side effects due to their common

off-targets implying a direct correlation between off-target binding and side-effect

similarity (Fliri et al. 2005). Thus, additional targets for FDA-approved drugs, often

implicated in entirely different therapeutic options and disease processes, can be

proposed. A method was developed to identify molecular activities of drugs that are

completely based on side effects but not implicit by their chemical similarity or the

sequence solely of their known protein targets (Campillos et al. 2008). The method

was able to identify alternative targets for many FDA-approved drugs, often

implicated in different therapeutic classes. The authors have used the relations

between side effect terms using Unified Medical Language System (UMLS) ontol-

ogy (Lindberg et al. 1993) to capture similarities between drugs. Finally, chemical

similarity is combined with side effect similarity to provide a final score for

assigning a probability to any pair of drugs to share a target.

4.3.3

Network-Based Approach

In this method, a comprehensive human protein-protein interactome was built from

fifteen commonly used resources with evidence from multiple types of experiments

(Cheng et al. 2018). Further, genes belonging to different types of cardiovascular

disease types were identified by Medical Subject Headings and Unified Medical

Language System vocabularies (Lindberg et al. 1993). For each cardiovascular

event, disease-related genes from eight frequently used databases were collected.

In addition, drug-target interactions on FDA-approved drugs from six frequently

used databases were assembled, and the interactions were weighted using reported

binding affinity data between drug and protein: inhibition constant, dissociation

constant, median effective concentration, or median inhibitory concentration.

Drug-target interactions were acquired from the DrugBank database (Wishart et al.

2018), the Therapeutic Target Database (Chen et al. 2002), and the PharmGKB

databases (Hewett et al. 2002). The bioactivity data of drug-target pairs were

collected from ChEMBL (Bento et al. 2014), BindingDB (Liu et al. 2007), and

44

S. Yellaboina and S. E. Hasnain