**2.1 Drug repurposing approaches**

At its broadest level drug repurposing approaches can be divided into two geneal types: computer-based and experimental techniques [20, 21]. Within both these approaches there are three main angles of attack, drug-centric, target-centric and disease-centric methods [22]. As indicated by its name, drug-centric approaches start from the point of view of a drug, with the aim to find efficacy against diseases other than the initial indication. In the case of a disease-centric approach the disease is the focus, with the purpose being to identify and repurpose a drug


*Abbreviations: MW, molecular weight; HBD, H-bond donor; HBA, H-bond acceptor; logP, octanol-water partition coefficient; clogP, calculated octanol-water partition coefficient; AMR, molar refractivity; PSA, total polar surface area; RGB, the number of rigid bond; HB, hydrogen bond; NAT, the number of atoms; NRTB, number of rotatable bonds; RNG, number of rings.*

#### **Table 1.** *Summary of 'druglikeness' rules applied in drug development.*

specifically against that disease. Target-centric methods utilize drugs that bind to well-characterized targets that are known to be, or at least suspected to be, involved in other diseases besides the drugs' original indication. What they have in common is that at the core these strategies often employ similarity assessment to identify drugs that can potentially be repurposed.

#### *2.1.1 Computer-based approaches*

Traditional drug repurposing often relies on the in vitro/in vivo identification of active drugs or alternative targets. Whilst this can provide promising compounds with reliable, desired activity, it can be expensive, involves physical access to the drug libraries and requires setup and optimization of the assays [23]. With the rapid development of bioinformatics and the accumulation of vast amounts of experimental data, the development of drug repurposing, especially the initial stage, has moved from traditional biological experiments towards an increasing diversity of computational screening approaches, partially due to the lower cost and lower barrier to entry [24].

#### *2.1.1.1 Virtual screening*

Virtual screening is an essential computational approach in drug discovery, and particularly in drug repurposing. It involves the use of computer programs to evaluate compound libraries on a specified criterion, usually similarity or calculated binding energy. Virtual screening can be classified into two categories: ligand-based and structure-based virtual screening [25, 26].

Ligand-based screening focuses on analyzing the structure-activity information of known active ligands against a certain indication to identify other potentially effective drugs. This analysis relies on similarity in the form of pharmacophores and geometric shape which can be informed by structural knowledge of the ligandtarget complex to identify key pharmacophores or without structural information, relying on the structure-activity relationship information from experimental approaches to identify the pharmacophores [27]. Pharmacophores are the chemical moieties of drugs that play essential roles in the interaction with their targets. Pharmacophore features—including features such as hydrogen bond donors, hydrogen bonds acceptors, charge groups, aromatic rings and hydrophobic centroids are then identified together with their spatial characteristics and mapped into a string [28]. This string can serve as a fingerprint, which can be used for easy similarity matching between different drugs, potentially identifying drugs that are also active against the disease.

This approach can be successful for small molecules as they are relatively simple molecules from a chemical perspective. Their pharmacophoric features are limited and usually rely on a few strong, deeply buried interactions with the target, making it easy to map and identify drugs with similar characteristics [29]. In contrast, biologics, such as peptides and antibodies, are far less suitable to these techniques as their method of actions typically dependent on mimicking proteinprotein interactions, which are characterized by large, flat interaction surfaces [30, 31]. This makes them highly specific for their target, allowing for targeted therapies with typically less side-effects, but that specificity also prevents them from being repurposed for a different target.

In contrast, structure-based virtual screening uses the three-dimensional structure of a target of therapeutic interest [32, 33], which is screened against a virtual library of approved drugs in order to identify those that show interactions with this novel target. Drugs are docked against the target and interaction analysis

#### *Drug Repurposing Techniques in Viral Diseases DOI: http://dx.doi.org/10.5772/intechopen.101443*

is performed based on binding energy and binding geometry. Many different docking software packages have been developed with the key differences being in the docking methodologies and the scoring functions used to rank the drugs [34]. Classical scoring functions usually rely on experimental data or prior information to rank the drugs. However, these have been consistently getting outperformed by machine learning based scoring functions, especially when specific target data is available to train on [35, 36].

Developments in computational power have made virtual screening approaches highly accessible to labs all over the world as they do not require nearly the amount of financial resources compared to wet-lab experiments. In addition, due to the speed at which the screenings can be performed nowadays, huge libraries containing 100's of millions of compounds can be screened against a target rapidly massively increasing the chemical space explored [37]. Though these advances are very useful in early drug discovery, where it can be used to screen fragment and compound libraries that cover a diversity in chemical space, they are less impactful when it comes to drug repurposing since the amount of approved drugs is limited and does not comprise wide chemical space. A downside of structure-based screening is the need for the actual structural information, which can be difficult to obtain for novel targets. Fortunately the number of entries available in the PDB has been growing at an exceptional rate, more than doubling in the last decade [38], meaning more and more targets have structural information available. In addition, the recent achievements of AlphaFold [39], including the prediction of the 3-dimensional structures of the entire human proteome, might alleviate this issue [40]. Overall this still has a positive impact on drug repurposing strategies as the more structural information is available the better scoring functions can be become, aiding both ligand-based and structure-based methods in identifying drugs that can be repurposed.
