Computational Approaches in Drug Repurposing

Abstract

Drug repurposing is a term applied to finding a new therapeutic and pharmacological indication for an existing drug molecule with a known indication. Repurposing existing drugs to treat both rare and widespread ailments is more and more compelling due to the use of less risky compounds, which may result in lower entire development costs and quicker development timelines. This is due to the high attrition rates, high cost, and slow new drug discovery and development pace. The introduction of computational techniques and their advancements in drug design, discovery, and development has provided a platform for scientists to kick-start drug repurposing with ease. Computational approaches have provided rationality in drug repurposing, reducing the chances of failure in drug repurposing attempts. In this chapter, we present techniques for drug repurposing that are both conventional and computational, talk about the difficulties faced by scientists who attempt drug repurposing, and suggest creative solutions to these difficulties to help drug repurposing reach its full potential.

1. Introduction

Drug repurposing simply means the science and technology of assigning new indications to exist molecules or medications with known therapeutic usage and safety profiles, most stemming from serendipitous discoveries [1]. According to the drug bank library of drug molecules, there are 4302 approved drugs [2, 3, 4, 5]. Though these drugs have been classified based on the target enzymes and pharmacological/therapeutic effects, they might still have the potential to activate or inhibit other enzymatic pathways, leading to different impacts on the body. Drug repurposing is all about utilizing and studying other possible enzymatic pathways or effects an already known drug can activate or inhibit, leading to pharmaceutical or pharmacological importance.

The traditional method of developing drugs is time-consuming and expensive; repurposing known drugs is a viable and promising alternative [6]. Developing a new drug involves studying its effectiveness, toxicity, pharmacokinetic, and pharmacodynamic profiles in cell- and animal-based investigations and its effectiveness and safety in humans in clinical trials. It typically takes 13 years and 2–3 billion dollars to develop a new drug from bench to bedside [7]. Drug discovery and development is a less attractive business for funding because of the rising costs and length of time. On the other hand, drug repurposing aims to identify new medical uses for an approved or experimental drug. Clinical trials can be hastened because the drug’s dosing and safety have been thoroughly investigated, considerably cutting the time and money needed for development [8].

Due to the high rates of illness and death associated with certain emerging diseases, repurposing drugs may be the most effective approach for addressing these conditions. When there is an urgent need to develop new medications and treatments during an outbreak, such as was the case with the COVID-19 pandemic, the strategy of quickly repurposing existing drugs has a significant advantage as it has the potential to identify medications that could be used to address the situation [9].

With current R&D costs, developing de novo drug therapies for more than 8000 rare diseases is inconceivable; nevertheless, drug repositioning, based on finding hidden associations or building connections between a drug and disease, holds promise for orphan drug disease therapy [10]. Furthermore, evaluating approved medications to determine new indications assists pharmaceutical companies in extending the patent life of drugs through application to adjacent diseases and in protecting IP against competitors [1].

Compared to the conventional drug development procedure, as shown in Figure 1, the advent of computation methods in medicinal research has provided a lesser expensive and less time-consuming approach to finding other disease conditions that can be treated using already approved or experimental drugs.

Figure 1.

Drug development vs. drug.

2. Repurposing principles

There are typically two main repositioning principles for drugs. First, because many diseases are interdependent, medications for one condition may also be effective for treating other disorders. Second, because medicines are naturally confusing, they can be linked to various targets and pathways. According to the source of the findings, drug repositioning research can be divided into two groups:

1. Drug-based tactics, in which discoveries are based on drug-related knowledge.

2. Disease-based strategies, in which discoveries are based on disease-related knowledge [9].

3. Repurposing techniques

In-silico approaches using existing data to find new drug-disease linkages and experimental screening approaches are the two main categories of systematic repurposing techniques.

3.1 Experimental screening approaches

Experimental screening approaches are used as a source of hits for drug discovery and drug repurposing, with notable differences in their application and outcomes. Searches in drug discovery programs are typically done for de novo candidate hits, fuelled by an HTS campaign, which requires highly specialized screening facilities and compound libraries containing several million compounds. Repurposing programs focus on advanced known molecules that either approved or failed with some knowledge of their safety or MoA available, led by in-depth screening and with smaller compound libraries. Typically approved compound libraries containing 500–2000 compounds and a similar number of existing but unapproved compounds are thought to be available.

3.2 In-silico approaches

In-silico repurposing techniques analyze data already in existence using sophisticated analytical methods to discover new possible connections between a drug and a disease [1]. The capacity to predict the conformation of small-molecule ligands inside the proper target binding site with a high degree of accuracy makes molecular docking one of the most commonly used in silico processes; after the creation of the first algorithms in the 1980s, molecular docking became a crucial tool in drug discovery. For example, investigations can be conveniently performed involving important molecular events, including ligand binding modes and the corresponding intermolecular interactions that stabilize the ligand-receptor complex. Furthermore, molecular docking algorithms execute quantitative predictions of binding energetics, providing rankings of docked compounds based on the binding affinity of ligand-receptor complexes [11]. Identifying the most likely binding conformations requires two steps: (i) exploration of an ample conformational space representing various potential binding modes; (ii) accurate prediction of the interaction energy associated with each of the predicted binding conformations. Molecular docking programs perform these tasks through a cyclical process in which the ligand conformation is evaluated by specific scoring functions [11].

In-silico approaches can broadly be divided into molecular techniques and real-world data approaches.

3.2.1 Molecular approaches

The molecular approaches are based on understanding drug activity and disease pathophysiology. They are often powered by large-scale molecular data known as omic data, including genomic, transcriptomic, or proteomic data and data based on drug targets and chemical structure. Due to the availability of datasets on drugs and diseases, as well as the robustness and reproducibility of the data, transcriptomics, and genomics are the two data types most widely used to support drug repurposing [12]. Transcriptomics studies the expression levels of thousands of genes, often accomplished by quantifying RNA using RNASeq or gene expression microarrays. One approach to using transcriptomics for drug repurposing is based on the idea that reversing gene expression signatures may result in a clinical benefit [1].

3.2.2 Real-world data approaches

The Real-world data approach focuses on identifying unknown and sometimes unexpected relationships between drugs and diseases or their symptoms. They are data based on individuals’ health, habits, and behavior captured without environmental intervention or bias introduced by data collection methodologies [1]. The real-world data approaches include network-based drug repurposing, ligand-based drug repurposing, structure-based drug repurposing, and machine-learning techniques [13].

3.2.2.1 Network-based drug repurposing

Network-based computational biology has become more prevalent in recent times. It integrates the relationship between biological molecules into networks to discover newly discovered properties at the network level and investigate how cellular systems induce different biological phenotypes under other conditions. A network can be represented as a connected graph in the network pharmacology framework, with each node representing either an individual molecular entity, its biological target, a modifier molecule within a biological process, or a target pathway, and each edge representing either a direct or indirect interaction between two connected nodes. An instance of this approach was demonstrated in 2009 by Hu and Agarwal, who utilized publicly available gene expression profiles from NCBI Gene Expression Omnibus (GEO) to construct a network that showed the similarity between different diseases. They then integrated this network with molecular profiles and knowledge of drugs and drug targets, which enabled them to identify opportunities for drug repositioning, as well as to propose molecular targets and mechanisms underlying drug effects [14]. In 2012, Jin et al. also devised a new method for repurposing drugs for cancer therapeutics that takes advantage of off-target effects that may affect critical cancer cell signaling pathways [15]. A hybrid model composed of a network component called cancer-signaling bridges and a Bayesian factor regression model was used to identify off-target effects of drugs on signaling proteins [13]. The main limitation of network-based approaches is that many biological aspects of the disease still need to be discovered, and network-based approaches may fail to produce promising drug candidates; also, biological elements interact with one another to form a complex system. As a result, this class of methods may have more practical effects [16].

3.2.2.2 Ligand-based drug repurposing

Ligand-based approaches are evaluated because similar compounds have similar biological properties. These methods have been widely used in drug repurposing to analyze and predict the activity of ligands for new targets. The number of publicly accessible compound records (more than a hundred million provided only by PubChem) is far greater than the number of deposited protein crystal structures (as of today, less than 150,000 in the Protein Data Bank) [17, 18]. Ligand-based methods rely on the chemical space coverage of already-known molecules. Deep learning and multi-task learning have been successfully used in ligand chemogenomic benchmark studies. When target and drug similarities were considered, the algorithm better predicted new drug-target associations. Machine-learning approaches play an essential role in silico Chemogenomics [13].

3.2.2.3 Structure-based drug repurposing

Structure-based similar protein structures increase the likelihood of performing similarly and identifying related ligands. Protein comparison is a technique used in medication repurposing to find secondary targets for a medicine that has already been licensed [19]. Proteins can be compared on a broad scale based on how similar their sequences are. The kinome is the most often-used example of a phylogenetic tree constructed using protein sequences [20]. In this tree, proteins from the same family are more likely to detect substrates or ligands that share similar functions, as in the case of dual inhibitors of the EGFR and ErbB2 receptors for an epidermal growth factor [21]. Sequence alignments work best when proteins have a high level of sequence identity. In contrast, local protein comparison works better when proteins share a low level of sequence identity to uncover unknown targets of known ligands [22]. It has become more crucial to compare protein binding sites to find local similarities [19]. This process is frequently followed by computing several descriptors that help determine a similarity score to locate cavities on the protein surface and compare binding sites. It is important to note that, when available, ligand binding modes are a valuable tool for finding new targets. Putting a focus on target-ligand interactions is one method of modeling molecular recognition. Several techniques, like structure-based pharmacophores or interaction fingerprints, can accomplish this. When the protein-ligand complex’s structure is unknown, one can predict hot spots in the binding site using computational approaches [23]. The viability of crystallographic structures of protein-ligand complexes is a prerequisite for structure-based techniques. The level of specificity that can be used to represent a binding site depends on resolution and sensitivity to atomic coordinates. While a protein’s static model can be seen in its crystallographic structure, conformational variations can cause the appearance of additional pockets [13].

3.2.2.4 Machine learning approaches

Although machine learning methods produce better prediction models, they are more data-dependent. Combining machine learning methods and other techniques can make an effective treatment plan for COVID-19 [16]. The general approach has been to fuse the structure-based and ligand-based screening methods with AI algorithms to build prediction models. AI and ML algorithms like deep learning, support vector machine (SVM), random forest (RF), Naive Bayesian, and neural networks have been extensively used for high throughput screening with lots of dataset molecules. In recent years, the development of next-generation computational methods using Artificial Intelligence (AI), Machine Learning (ML), and network medicine approaches has positively impacted the different stages of drug development [24].

4. Success stories in computational drug repurposing

The field of data science is blooming, and its role in detecting potential candidates for drug repurposing has yet to be explored. There are various approaches to drug repurposing, but the computational approach is unique in the way it utilizes neither in-vivo nor in-vitro techniques. It is known as in-silico drug repurposing—an expediting, cost-friendly, and reliable process [25]. This method relies heavily on data from diverse sources like electronic health records (EHRs) comprising disease diagnoses, lab test results, medical prescriptions, genetic data from biobanks, chemogenomic data, and proteomic data [26]. These data sources, when collated and analyzed, are then capable of producing valuable insights. A few instances:

Given widespread tuberculosis and its extensive resistance mechanisms to current anti-infective treatment, Kleandrova et al. performed a study on computational drug repurposing for antituberculosis therapy by creating a multi-condition model based on quantitative structure–activity-relationship (QSAR) [27]. This sought to find potential antituberculosis agents capable of acting as inhibitors of multiple strains of the bacteria. The model utilized a combination of perturbation theory concepts and machine learning techniques to screen large data repositories for chemical structures with the potential to inhibit Mycobacterium tuberculosis, the causative organism. The dataset comprised 8898 agency-regulated chemicals, including investigational and FDA-approved drugs. After that, stipulated metrics were used to rank these agents, with priority given to those exhibiting the highest values. Top of the list was macozinone, BTZ-043, and niclosamide, but niclosamide is a popularly known anti-helminthic. This drug is believed to have anti-parkinsonian, anti-diabetic, and anti-viral properties [28]. It is also important to mention that through computationally identifying drugs that can increase the mRNA expression of downregulated genes in hepatocellular carcinoma (HCC) and decrease the mRNA expression of upregulated genes, the antitumor activity of niclosamide and its ethanolamine salt (NEN) was discovered. The antiproliferative activity of niclosamide and NEN in different HCC cell lines and primary human hepatocytes was then evaluated in vitro. This was further confirmed by in vivo testing against two mouse models (genetically induced liver tumors and patient-derived xenografts [PDXs]) for HCC to show a substantial reduction in the cancer progression after oral administration of NEN compared to niclosamide [29].

Similarly, Zhang et al. performed thorough data mining to identify drugs with anti-Alzheimer properties [30]. Their study revealed seven drugs inhibiting acetylcholinesterase, a known drug target of most anti-Alzheimer conventional medicines. These drugs, which have never been used in the management of Alzheimer’s, can be used in the future for cognitive deficiency therapy in patients with the disease. Zhang et al. previously conducted an identical study for drugs that can be used for anti-diabetic treatment [31]. Using data mining and pathogenesis information, their study repurposed 58 drugs, out of which nine were prioritized for having higher potential in treating diabetes. Among these nine drugs were four (diflunisal, nabumetone, niflumic acid, and valdecoxib) used in rheumatoid arthritis, osteoarthritis, and pain management. Connectivity map analysis showed that cells treated with these four drugs had similar gene expression as cells treated with conventional anti-diabetic medications like metformin and glimepiride. Evidence from Koren et al., 2019 also suggests that a different class of drugs, the alpha-1 adrenergic antagonists, might have a potential impact on diabetes control [32]. These success stories, though sparse in their numbers, hold a promise for the future. Diseases like diabetes often last for a lifetime, and an estimated 400 million people [33] worldwide suffer from it; therefore, integrating the results of this expediting approach to drug discovery into clinical practice will revolutionize modern medicine.

5. Limitations

Drug repurposing by pharmaceutical companies faces many challenges. There is a need to create a business model to support the use of existing molecules as therapeutics for new indications and repurposing drug pathways. There is also a need to demonstrate the effectiveness and recover the investment required to bring recycled products to market [1]. Furthermore, this methodology is based on structural files and cannot be used immediately when identifying a new or orphan target [24]. This is because a more extensive collection of records may not be achieved since there is no defining identifier to connect data [16]. This can be seen in the Artificial Intelligence, Machine Learning, and network medicine approaches of computational drug repurposing, which require large amounts of data to train models. Lack of access to structured, standardized data related to analytics and clinical trials can impair the tool’s predictive ability. Furthermore, the majority of developed models are local models; that is, they are specific to one problem, and there is no global model or suite that helps in solving or querying the wide range of problems drug discovery teams may encounter [24].

All computational-based drug repurposing methods heavily depend on data. Existing databases pose lots of challenges for researchers. The volume of data in some databases needs to be increased to generate a suitable model, and there is no determinant identifier to connect data to collect more comprehensive datasets. Data descriptions could be clearer, making it easier to understand them. The databases also contain data for a specific purpose rather than complete data. Lastly, introducing new Active Pharmaceutical Compounds (API) commands has made them difficult to learn and use. Existing databases have some limitations that can be overcome using software engineering techniques [16]. In terms of improving efficacy and reducing the time and cost of a drug discovery project, computational-based approaches may produce more acceptable results than others. Every computational drug repurposing method has advantages and disadvantages and heavily depends on data [16].

The computational approach is auspicious and effective in other domains. Natural language processing, for example, has proven helpful in translation, spell-checking, and other applications. However, AI/ML-based techniques necessitate a large amount of data to train the models. The inaccessibility of structured and standardized data associated with assays and clinical trials may jeopardize the tools’ predictive ability. Furthermore, most developed models are local, which means they are specific to one problem. No global model or suite can help resolve or query a wide range of issues that a drug discovery team may frequently encounter [24].

6. Opportunities in the computational repurposing of drugs

Although sciences and technology have progressed rapidly, de novo drug development has been costly and time-consuming over the past decades. Given these circumstances, “drug repurposing” (or “drug repositioning”) has appeared as an alternative tool to accelerate the drug development process by seeking new indications for already approved drugs rather than discovering de novo drug compounds, nowadays accounting for 30% of newly marked medications in the U.S [34]. Even though the application of computational methodologies to drug repurposing has yielded some positive results and has been propounded to repurpose drugs on a large scale by utilizing available high-throughput data, due to the failure of the current drug regimen, many more diseases need urgent attention in terms of new drug therapies. There are increasing number of deaths from Neglected tropical diseases. The World Health Organization (WHO) describes neglected tropical diseases (NTDs) as a diverse group of communicable diseases that prevail in tropical and subtropical conditions [35]. Neglected Tropical Diseases include Buruli ulcer, Chagas disease, dengue and chikungunya, dracunculiasis (Guinea-worm disease), echinococcosis, foodborne trematodiases, African human trypanosomiasis (sleeping sickness), leishmaniasis, leprosy (Hansen’s disease), lymphatic filariasis, mycetoma, chromoblastomycosis, and other deep mycoses, onchocerciasis (river blindness), rabies, scabies, and other ectoparasitoses, schistosomiasis, soil-transmitted helminthiases, snakebite envenoming, taeniasis/cysticercosis, trachoma, and yaws and other endemic treponematoses [35]. According to WHO, NTDs cause about 200,000 deaths yearly [35]. A person may become severely disabled, disfigured, blind, or malnourished after contracting an NTD and frequently acquire multiple NTDs at once. If new drugs have to be developed for these conditions through conventional means, many deaths must have been recorded before the drugs get to market.

According to Nigeria Centre for Disease Control (NCDC), in 2021 and 2022, Cholera killed more people in Nigeria than COVID-19 [36]. Even though there are standard treatment guidelines for this condition, the death rate keeps rising. Globally, lives are being lost from different types of cancers, even with all the treatments currently available. Lives are also being lost from various other diseases affecting mankind. Computational drug repurposing will go a long way in providing within a short time a possible better treatment and management options for these diseases and all other diseases challenging mankind.

7. Conclusion

The utilization of existing drugs to identify other potential therapeutic indications can be done more quickly and with less expense through computational drug repurposing. This approach is facilitated by the use of protein and chemical databases, which have been developed to support computational techniques. These databases enable existing drugs to be acquired in the necessary formats for computational studies. There is now a broad selection of available computational tools, with more currently under development, which can help to advance the application of computational approaches in drug repurposing. With access to these tools and databases, any researcher with an interest in this area can begin to explore drug repurposing. The effective use of computational drug repurposing has the potential to improve treatment and management options for a wide range of diseases affecting humanity.

Acknowledgments

The authors express their gratitude to the CURIES research group, whose support and encouragement have been invaluable.

Conflict of interest

The authors declare no conflict of interest.