Molecular Docking: An Insight from Drug Discovery to Drug Repurposing Approach

Abstract

The impact of computer-aided drug designing in the field of medicinal chemistry has created a boon in the drug discovery process. Molecular docking is an integral part of bioinformatics that deals with protein-ligand interactions, binding conformations, and affinity predictions. It has shown to be a rapid, easy, and affordable method in business and research settings alike. The advancement in the hardware and software has led to enhanced molecular dynamic simulations and thus authenticate the computational results. This has created a great impact in minimizing the cost and time involved in the drug discovery process. It has also helped in identifying the rationale for drug repurposing approaches. This chapter will give in-depth knowledge of the importance of molecular docking in drug designing and discovery, their impact on drug repurposing, and success stories of the in silico approach in drug discovery and repurposing.

1. Introduction

As estimated by a study in 2016 by the Tufts Center for the Drug Development Study, the cost of a new drug development till the marketing step has increased by almost 144%, which sometimes does not go well if the drug gets recalled [1]. De novo drug development is a difficult, time-consuming, and costly procedure that needs a significant financial commitment. It can take anywhere from 13 to 15 years and 0.8 to 3 billion dollars on average to develop a medicine from a theoretical hypothesis to a new commercial medicinal molecule [2, 3, 4, 5].

Drug research and discovery typically entail four steps of preclinical testing and clinical trials. According to estimates, the Food and Drug Administration (FDA) evaluates around nine out of ten drugs each year. Only 10% of the medications enrolled in phase 1 clinical trials are authorized; the remaining treatments fail because of severe toxicity or ineffectiveness. The primary cause of these attritions is incorrect medication target or response identification. Therefore, searching for the correct lead compound is crucial to the project’s overall success in the difficult process of drug discovery. This not only contributes to reducing the economic burden of society but also manages the timeline associated with dealing with the morbidity or mortality of the disease as evident from the COVID-19 scenario.

2. Drug designing and discovery

Drug development is a long, arduous process that costs the galaxy. Preclinical testing is typically the first step in the four-phase clinical evaluation process that follows drug discovery and development. It has historically changed over the ages in step with advances in science and technology. Historically, the drug development approaches consist of Serendipity where drugs’ original activities were found accidentally, as was the case of penicillin discovery [5]. A synthetic and semi-synthetic approach is related to the advancement in the field of chemistry. Replacing the natural origin compounds with synthetic or semi-synthetic alternatives reduces the cost of the product and meets the needs of the population. Off-label use means that they were being used for purposes different than those for which they were initially licensed [6]. For instance, propranolol was initially approved in the US in 1968 for the treatment of arrhythmias; but, 10 years later, it was also approved for the treatment of angina pectoris and antihypertensive medication [3].

Consequently, the goal of drug development is to create pharmaceuticals that bind to a certain protein target more firmly than the natural substrate. This allows for the modification or inhibition of the biological reaction that the target molecule catalyzes. Typically, drugs are found by accident through a process of trial and error utilizing high-throughput screening techniques that test the activity of several compounds against a target in vitro. This is a costly and time-consuming process [7, 8].

Simulated molecular docking can be a helpful technique in the drug-discovery process if the target’s 3D structure is known. By virtually screening compound databases, this in silico method enables the identification of promising therapeutic candidates more quickly and affordably. Once the drug candidates found by the virtual screening procedure have been discovered, additional research can be done on them through toxicological testing, clinical trials, lab experiments (synthesis), and other means Figure 1 [9].

Figure 1.

Conceptual diagram of de novo drug development and drug repurposing. Drug repurposing starts with target discovery for an already-approved medication, which is immediately followed by phase 2 and 3 clinical trials. Phase 1 clinical studies and animal studies were not carried out because the information from these studies is already accessible for an already-approved medication [4, 7].

3. Drug repurposing strategies

Considering these challenges, it can be profitable to look for new uses and targets for pharmaceuticals that are already on the market. This strategy, known as “drug repurposing” (also known as “drug repositioning” or “drug re-profiling”), was initially presented by Ashburn and Thor in 2004 to find treatments for diseases for which there are none. In technical words, the way to identify the novel medication indications has already been approved. Such circumstances can include uncommon illnesses or a pandemic that is spreading quickly and for which there is no treatment [3, 4, 5]. There are two general repurposing approaches illustrated in Figure 2.

Figure 2.

Schematic illustration of repurposing approaches.

3.1 Knowledge-based repurposing

With this repurposing approach, models are created to forecast unidentified targets, biomarkers, or disease mechanisms by using drug-related data, such as drug targets, chemical structures, pathways, and side effects [10]. Target-based entails in silico screening of pharmacological compounds from drug libraries (Drug bank, PubChem, Zinc database, ChemBl, etc.), for ligand-based screening or docking, after high-content screening and/or high-throughput (HCS/HTS) of medicinal compounds given proteins or biomarkers of interest [11, 12]. The target-based strategy has the benefit of screening almost all therapeutic molecules with known chemical structures. Pathway-based medication repurposing makes use of data from protein-interaction networks, metabolic pathways, and signaling pathways to forecast the relationship or resemblance between a drug and a condition (KEGG, MetaCyc, FragariaCyc, etc.) and target mechanism-based drug repurposing. Target mechanism-based drug repurposing integrates treatment omics data, protein interaction networks, and signaling pathway information to discover new mechanisms of action for drugs (GenBank, UniProt, SWISS-2D PAGE, NCBI databases, etc.) [13]. These repurposing approaches have the benefit of seeking to understand the mechanisms pertaining not only to pharmaceuticals or diseases but also to medication therapies for disorders.

3.2 Signature-based repurposing

Gene signature data from disease omics data are utilized in signature-based repurposing to identify novel off-targets or disease processes. To identify inverse relationships between drugs and diseases, this strategy examines the gene expression profiles of the medication and the illness. These methods have the benefit of revealing novel pharmacological mechanisms of action. Additionally, these strategies entail more molecular and/or genetic-level mechanisms than knowledge-based methods do [4, 14].

3.3 Phenotype-based repurposing

The availability of phenotypic data has opened new avenues for medication repositioning. System techniques have been using these kind of data more and more in the past few years to identify genetic features linked to human illnesses (Phytozome, PubMed, Gramene database, EnsemblPlants, etc.) [15].

4. Molecular docking studies

Molecular docking techniques play a major role in the design and development of novel medications [16, 17, 18] by predicting the experimental binding mechanism, affinity, and location of a small molecule (ligand) within the binding site of the target receptor (macromolecule) [19, 20]. The docking process (Figure 3) can accurately predict the native ligand posture at the receptor binding site (i.e., the experimental ligand geometry within a given tolerance limit) and the corresponding physical-chemical molecular interactions. It must also be able to accurately rank these ligands among the top compounds in the database and reliably distinguish binding from nonbinding molecules while analyzing large compound libraries [21, 22].

Figure 3.

Molecular docking process.

The primary step in the docking process is agreeable spatial and energetic matching between the ligand and receptor to obtain the optimal conformation. Furthermore, when the ligand attaches to the receptor, the structure of the binding site changes continually until a stable connection is correctly created. Only then can the configuration and charge distribution of the binding system be determined [16].

Pre-organization and geometric complementarity-based energy complementarity ensure that ligands and receptors have the most stable structures while consuming the least amount of free energy. The molecular docking program can help us find the optimal shape and orientation in terms of complementarity and pre-organization by employing a specific technique. Next, we may predict the binding affinity and look at the interactive mode using a scoring formula. Molecular docking is a widely utilized technique in the computer-aided structure-based drug development era to ascertain the anticipated geometry of a protein-ligand interaction [9].

Automation, parallelization, flexibility, sampling, optimizations, scoring, and posture prediction are all part of the docking process. Improvements over widely used algorithms are almost usually claimed: quicker docking, more precise affinity prediction, or more engaging graphics [21, 23]. A search algorithm and an energy scoring function are the basic tools of a docking methodology for generating and evaluating the ligand conformations [21]. Various molecular docking algorithms can be employed to predict protein-ligand configurations and order them based on the scoring functions integrated into every distinct docking approach [23, 24]. For molecular docking, the most popular heuristic search methods have been evolutionary, tabu, and simulated annealing [25].

To ascertain the most advantageous ligand conformation when bound to the target, docking techniques typically utilize an energy-based scoring system. A variety of scoring functions are available, including empirical, knowledge-based, force field-based, clustering and entropy-based, and consensus scoring approaches. The presence of water molecules at the active site is another aspect of docking target flexibility. It is generally agreed that lower energy scores represent better protein-ligand bindings than higher energy values. As a result, molecular docking can be formulated as an optimization problem where the objective is to determine the lowest energy ligand-binding mode Table 1 [22, 25].

S.No.	Molecular docking approaches	Working principle
1	Fragment-based method	Fragment-based methods include breaking down the ligand into separate photons or particles, attaching the fragments, and then joining the fragments.
2	Inverse docking	When combined with a precise pharmacokinetic property, knowledge of all these targets can help assess a drug candidate’s potential for toxicities and side effects. To conduct docking studies on a particular ligand, a special protocol is selected.
3	Blind Docking	This method was created to scan the entire target molecule interface to find putative peptide ligand binding sites and mechanisms of action.
4	Point complimentarily approach	These methods concentrate on contrasting the compositions and/or geometries of various molecules.
5	Matching approach	It places a strong emphasis on redundancy and determines the best place for the ligand atom within the site, which may lead to a ligand-receptor combination that still needs work.
6	Ligand fit approach	It is rapid and accurate method for docking tiny ligands into protein active sites while accounting for shape complementarity is called ligand fit.
7	Monte carlo approach	It causes a ligand’s translations, rotations, and conformation to change randomly within an active site. An initial configuration value is assigned by it. After that, it evolves and generates a fresh configuration. The Metropolis criterion is used to determine whether the new configuration is maintained.
8	Distance geometry	It is possible to express several types of sequence properties in terms of intra- or intermolecular dimensions. These distances can be assembled and three-dimensional structures that work with them can be computed according to the framework of distance geometry.

Table 1.

Different molecular docking approaches [9, 23, 26].

5. Models of molecular docking

Emil Fischer’s lock-and-key assumption in 1894 [8, 16, 27] states that both the ligand and the receptor may be thought of as rigid bodies and that their affinity is directly proportional to a geometric fit between their shapes, served as the foundation for the earliest known docking techniques. Subsequently, Koshland’s “induced-fit” idea advocated that the ligand and receptor should be viewed as flexible during docking [8, 16, 28].

Unlike relatively autonomous side chains, each movement of the backbone affects numerous side chains. As a result, compared to flexible docking with a stiff receptor, the sampling process in a fully flexible receptor/ligand docking has a higher order of magnitude of degrees of freedom. As a result, these flexible docking algorithms can anticipate a molecule’s binding mode and binding affinity concerning other compounds with greater accuracy than rigid body algorithms. The conformation ensemble model works beyond small induced-fit alterations, and proteins have been found to experience much larger conformational changes. A novel idea says that proteins are made from an ensemble of conformational states that already exist. Because of its flexibility, the protein can change states [26].

6. Types of docking

A technique for creating, characterizing, and altering compound topologies and interactions as well as the properties of these molecules that depend on their three-dimensional geometries is molecular modeling [26]. During the last 20 years, the development of molecular simulation (MS) technology has exhibited significant progress, especially since the 2013 Nobel Prize in Chemistry was awarded. As of right now, free energy calculation, molecular docking, homology modeling, and molecular dynamics simulation are all included in MS technology. The most popular approach among these in molecular modeling research is molecular docking [16, 29, 30, 31, 32, 33]. Generally molecular docking can be Automated or manual. The ligand is matched with its complementary group in the binding site during manual docking, where binding groups on both the ligand and the binding site are known. Each possible interaction’s bonding distance is specified. To find the optimal fit as specified by the operator, the program shifts the molecule about inside the binding site. The matched groups are fitted so that they are within the desired bonding distances of one another rather than being directly superimposed.

It is possible to perform automatic docking, in which the software determines how to dock the ligand. The task for the docking program is twofold. I: it must arrange the ligand in various binding modes or orientations within the active site. II: to determine which binding modes are optimal, it must score each one. The order of complexity may be (a) target and ligand as rigid bodies; (b) ligand as a flexible body and target as a rigid substance; and (c) both target and ligand as flexible body.

The goals of docking simulations require, there are several types (Figure 4) of molecular docking processes that involve either flexible or stiff ligand/target combinations [21, 34, 35]. In Rigid docking, we assume that the compounds are stiff, and thus want to find a strategy to rearrange one of the compounds in three dimensions so that, in the context of a scoring system, it best fits (Figure 5) [23, 28, 29, 36]. It is a good fit for large systems like protein-protein and protein-nucleic acid. It is the simplest approach as it does not need several calculations [37, 38].

Figure 4.

Flow diagram of molecular docking types [8].

Figure 5.

Lock and key model [26].

The flexible docking approach facilitates easy conformation modification of the docking system (receptor and ligand) during the process. As the variables of the receptor and ligand rise with the number of atoms, several other parameters must be considered. These include the computation’s significant complexity and the docking procedure’s excessive complexity. To effectively analyze the interaction between molecules, flexible docking is typically utilized [16, 39]. We assess molecular flexibility in conjunction with transformation to find confirmations for the ligand and receptor molecules as they are present in the complex (Figure 6) [23, 26, 36, 40].

Figure 6.

Induced-fit model [26].

Another approach is semi-flexible docking where the ligand’s conformation changes during the semi-flexible docking process, but the fixed receptor’s conformation stays the same. Consequently, the noncritical portion of the ligand structure’s bond length or angle can be fixed [41]. For the docking of macromolecules like proteins or nucleic acids as well as tiny ligand molecules, semi-flexible docking is appropriate [42]. It can be used in a wider range of applications and considers the impact of ligand structural modifications [16]. Ensemble docking is more specific for protein interactions which clarifies the intricacy and flexibility of protein structural states. An ensemble of protein structures is used to bind with a ligand. Ensemble docking (Figure 7) is a potent technique for discretely representing the flexibility of a target receptor by docking molecules into several or multiple conformations of the receptor. This method offers a structural degree of freedom, where any ligand may find a compatible receptor, imitating the dynamic activity of the protein. In actuality, the latter is dependent upon the conformational space that the ensemble occupies. The general assumption is that ensemble docking is better than docking into a single receptor conformation [23, 40, 43, 44].

Figure 7.

Ensemble docking [21].

7. Methods in molecular docking

Molecular docking turned out to be a powerful tool in drug discovery. The first molecular docking algorithm was introduced in the 1980s and later methods in molecular docking have evolved drastically. Now, it is possible to analyze important molecular events such as binding conformations of the ligand and its interactions with the receptor that alleviate the ligand-receptor complex. Modern molecular docking algorithms exhibit the potential to quantitatively envisage binding energetics to categorize the docked ligands based on the binding energies of their complexes with the receptors. This section will introduce different existing molecular docking methods.

7.1 Conformational search methods

The basic goal of conformational search methods is to recognize the energetically favorable binding conformations of the ligand-receptor complex. Conformational search methods tend to incrementally amend ligand’s torsional, translational, and rotational degrees of freedom by using either systemic or stochastic techniques.

7.1.1 Systematic methods

Systemic methods tend to perform an exhaustive and ordered search for the conformational landscape of the ligand by systematically varying its degrees of freedom. This method identifies all the combinations of structural constraints. Depending on the degrees of freedom of a particular ligand, several combinations are made that leads to a phenomenon called combinatorial explosion.

Surflex-Dock, Dock, FlexX, SLIDE, GLIDE, FRED, ADAM, EUDOC, Ehits, Flog, and Hammerhead are examples of Systematic methods [17].

7.1.2 Stochastic\random methods

Genetic algorithm (GA) is an exciting application of stochastic search algorithm where conformational optimization is carried out through the process of natural selection and genetics. It is used to find out a proper ligand’s conformational space along with its energetically favorable binding modes. This algorithm is mainly used while working with flexible ligands or complex binding pockets [17].

Genetic algorithm begins with a preliminary population of different ligand conformers which are randomly generated. Using a scoring function that usually considers certain parameters like hydrogen bonds, hydrophobic contacts, salt bridges and steric hindrance, the fitness of all the generated ligand conformers is assessed. Based on the principle of “survival of the fittest,” ligand conformers (also known as parent ligand conformers) showing better fitness scores are selected for the next round, that is, recombination and mutation. Whereas, torsional angles and other conformational parameters from two parent conformers are merged followed by their modest alterations in order to identify new regions of conformational space. This new population then enters another round of recombination and mutation to generate more improved ligand conformers. This process continues until the fulfillment of the required criterion where an acceptable fitness level is obtained.

Autodock, Gold, LigandFit, Molegro Virtual Docker, PLANTS, GlamDock, and MOE_Dock are examples of stochastic methods [17].

7.2 Energy scoring functions

To approximate the binding energies of the predicted ligand-receptor complex, certain scoring functions are used. These scoring functions normally rely on a few physicochemical parameters including intermolecular interactions, entropy, and desolvation [45]. There is an indirect relationship between the number of physicochemical parameters evaluated and the productivity of the docking algorithm because the computational cost increases with an increase in number of variables. An ideal scoring function should maintain a balance between score and speed.

Different scoring functions have been categorized into three different classes including force-field-based, empirical, and knowledge-based scoring functions [46].

7.2.1 Force-field-scoring function

This de novo method is applied to estimate the binding energy by generally adding the contributions of bonded and non-bonded terms [47]. However, due to the lack of realistic physical models for describing entropic effects and desolvation, the usage of this scoring function is limited [48]. Dock, AutoDock, and Molegro Virtual docker are well-known docking methods that utilize this scoring function.

7.2.2 Empirical scoring functions

These functions are derived from empirical observations and experimental data involving ligand-protein complexes with known binding energies rather than from first principles or physical equations. Due to the simplified way of estimating binding energies, empirical scoring functions are faster than force-field-based scoring functions. However, their dependence on the precision of data used to develop the training model is a major limitation. AutoDock, ChemScore, Surflex, and GlideScore are popular molecular docking method that utilizes empirical scoring functions [49, 50].

7.2.3 Knowledge-based scoring function

In knowledge-based scoring functions, statistical information retrieved from experimental structures is leveraged for examining the possibility of molecular interactions. Additionally, these scoring functions also consider the preference and occurrence of certain molecular interactions in the experimental data.

As these scoring functions neither calculate the binding energies de novo like fore-field scoring nor reproduce binding energies like empirical scoring, they present a balance between accuracy and speed. Molecular docking software that uses these scoring functions are SMoG, RF_Score, DrugScore, etc.

7.3 Covalent docking

Typically, drug designing protocols tend to search non-covalent ligands/inhibitors to avoid the toxicity problems of covalent inhibitors. However, during the SARS COVID-19 outbreak, the role of covalent drugs has been appreciated in terms of better efficacy compared to the native substrate of their targets.

Covalent docking is used to genuinely model covalent interactions between the ligands and receptors as it creates a tight bond between the electrophile (ligand) and nucleophile (protein), and it can produce chemical probes with high levels of potency and selectivity. This procedure offers the investigation of irreversible binding events that lead to the analysis of the crucial covalent drug interactions. This type of docking is mainly important in designing targeted therapies where irreversible ligand binding is required for improved efficacy and selectivity. However, the designing of the covalent ligand is somewhat difficult as certain molecular events including the creation and breaking of covalent bonds along with their rearrangements must be handled with Quantum Mechanics (QM). Notably, QM cannot adequately be operated by the traditional force fields or empirical techniques of molecular docking [51]; however, the requirement of QM can be avoided using faster and uncomplicated molecular modeling techniques. Fewer molecular docking methods have tackled the problem of covalent bond modeling including DOCK, GOLD, and AutoDock [17]. Numerous FDA-approved medications, including aspirin, warfarin, azacitidine, isoniazid, and others, have been discovered to exhibit covalent bonding [21].

7.4 Docking considering water molecule

Ligand-receptor interactions are simulated within the living system; certain docking algorithms have been introduced recently that evaluate the involvement of water molecules along with the ligand within the active (binding) site. The objective of considering the solvent setting improves the prediction accuracy, mostly in cases where water molecules play an important role in stabilizing the ligand-receptor complex.

In the Protein Data Bank, the crystal structures of more than 59% of the protein-ligand complexes were shown to contain at least one water molecule in their active site forming multiple hydrogen bonds between the ligand and the active site of the protein. These active site water molecules can be dislodged with the help of synthetic ligands or taken as a portion of the receptor protein. Certain methods are available to calculate the binding free energy for a given water molecule and, thus, distinguish between the dislocatable and structural water molecules [52].

Monte Carlo simulations, GOLD, AutoDock, HINT force field, etc., can perform molecular docking investigations considering structural water [53]. Nonetheless, there is a noticeable disparity in accuracy and capability across the many docking software versions that are currently on the market. In the literature, around 59 molecular docking tools have been published over the past 20 years for usage in academia and business, including numerous others.

8. File formats used in molecular docking

Different molecular docking programs rely on specific file formats of protein receptors and ligands. These file formats offer a standardized way to represent molecular structures of ligands and receptor proteins, facilitating the harmonization of different molecular docking software. The selection of a specific file format often depends on the specific requirements of the docking software being used. Some of the important file formats used in molecular docking are described here.

8.1 MOL2 (Tripos Mol2)

A Tripos Mol2 file (. mol2) is an ASCII file that is too absolute and manageable to represent a SYBYL molecule. Moreover, it has all the information required to rebuild a SYBYL molecule. Contrary to other fixed format files, Tripos Mol2 file is made in a free format that can be easily converted to another file format. Conventionally, MOL2 files describe the atom types, three dimensional coordinates of each atom, partial charges, types of bonds etc. The MOL2 format is frequently used to represent the three-dimensional structures of the ligands. Many molecular docking software requires ligand structure file in MOL2 format.

8.2 SDF (structured data file)

The structured data file (SDF) is a type of chemical-data file format that was designed by Biovia formerly known as Molecular Design Limited (MDL). The SDF provides two-dimensional or three-dimensional structural information of the ligands in plain text. Contrary to the MOL file which usually contains the structural information of one ligand, the SDF file format can hold single or multiple ligands separated by the sign of four dollars ($$$$). Additionally, SDF files also encode data on connectivity and hybridization state. SDF files are commonly used to represent ligand structures, especially in virtual screening studies.

8.3 PDBQT

The PDBQT (Protein Data Bank, Partial Charge, and Atom Type) format is considered an extension of the PDB (Protein Data Bank) file format. Overall, the PDBQT format is like the PDB format; however, it contains a bit more information for each of the ligand’s atoms especially regarding their partial charges (q) and atom types (t).

The calculation of partial charges (q) is considered central for calculating the electrostatic interactions between ligands and receptors during the docking process carried out by the AutoDock tool. However, Autodock Vina does not rely on these charges for computing binding affinities. Rather, Vina considers the atom type (t) as a crucial variable in estimating the binding energies. It evaluates the role of atom types in ligand-receptor interactions as atom types are important in describing their behavior in the molecular docking process including interaction potentials and steric hindrances. PDBQT format can also be used to depict flexible compounds as it can also incorporate information about rotatable bonds so that the docking software can investigate the conformational space of the ligand during the docking process. In AutoDock-based molecular docking studies, PDBQT format is required for both the ligand and the receptor files.

8.4 XYZ (Cartesian coordinates)

The XYZ file format is known for its simplicity as it represents the total number of a ligand’s atoms in the first row and a commentary in the second row. From the third row, different atoms are enlisted with their three-dimensional or Cartesian (X, Y, and Z) coordinates. XYZ files can be easily generated to be used in some docking software, especially for small ligands [54].

9. Active site prediction

The active site or binding site of an enzyme refers to the location where the substrate is fixed, which is typically situated on the surface of the enzyme. The term “ligand-binding region” refers to a segment of a protein or DNA/RNA molecule that is responsible for forming bonds with chemicals, drugs, or ions. This interaction results in a conformational change, ultimately leading to physiological, agonist, or antagonist effects [55]. The ligand-binding region is often characterized by a three-dimensional cleft or cavity that is shaped by the arrangement of amino acids in sequence.

According to recent research, it has been discovered that the central segments of numerous proteins have been nearly perfected for stability through natural selection. On the other hand, surface residues have not been optimized to the same extent, likely because protein function is facilitated through surface connections with other molecules. The proliferation of structural genomics initiatives has resulted in a substantial rise in the number of accessible protein structures. However, a significant portion of these structures are designated as hypothetical proteins due to the absence of biochemical information.

The experimental characterization of protein function is a labor-intensive and time-consuming process. To address this challenge, a computational tool for predicting functional sites in proteins would be invaluable. The value of such a tool is further emphasized by the automation required for structural genomics projects [56].

Based on the sequence or structural similarity of the query protein with well-characterized proteins, a variety of theoretical techniques are available to try and predict the activities of proteins. Proteins with identical sequences or structures may not always carry out comparable biological tasks. Similar tasks are also known to be carried out by proteins with distinct folds, such as subtilisin- and trypsin-like proteases [57]. Because of this disparity, structure-based methods have been developed, in which the similarity of the spatial arrangement of functionally important residues is used to predict function.

As a result, there is now more focus on determining the active site because it will impact the ligand’s function and reaction. Energy functions, solvation models, and sampling algorithms capable of high-resolution structure prediction are necessary because binding and active sites are solvated or contain several polar and charged residues that create directed contacts to their ligands [58].

Following protein production, it is necessary to anticipate the protein’s active site. Even though the receptor may have many active sites, only one of them needs to be selected. If heteroatoms and water molecules are present, they are mostly eliminated [59]. For carrying out molecular docking, we must be aware of the active site, either from literature research or active site prediction. We have no choice but to use the active site prediction if the literature does not provide sufficient evidence. Numerous online web servers are available to assist in predicting the active site, and some of them are listed below in Table 2.

S.No	Servers	links
1	CASTp	http://sts.bioengr.uic.edu/castp/
2	LigASite	http://www.bigre.ulb.ac.be/Users/benoit/LigASite/index.php?home
3	PDBeMotif	http://www.ebi.ac.uk/pdbe-site/pdbemotif/
4	fPOP	http://pocket.uchicago.edu/fpop/
5	metaPocket	http://metapocket.eml.org/
6	PocketQuer	http://pocketquery.csb.pitt.edu/
7	IBIS	http://www.ncbi.nlm.nih.gov/Structure/ibis/ibis.cgi
8	3DLigandStie	http://www.sbg.bio.ic.ac.uk/~3dligandsite/
9	SitesBase	http://www.modelling.leeds.ac.uk/sb/
10	PRO SURFER	http://163.42.136.93/top
11	KBDOCK	http://kbdock.loria.fr/index.php
12	Pocketome	http://www.pocketome.org/
13	sc-PDB	http://cheminfo.u-strasbg.fr:8080/scPDB/2011/db_search/about_scpdb.htm
14	ProBiS	http://probis.cmm.ki.si/index.php
15	Active Site prediction	http://www.scfbio-iitd.res.in/dock/ActiveSite_new.jsp
16	INDSITE	http://cssb.biology.gatech.edu/findsite
17	PocketDepth:	http://proline.physics.iisc.ernet.in/pocketdepth/
18	GHECOM 1.0	http://strcomp.protein.osaka-u.ac.jp/ghecom/
19	Pocket-Finder	http://www.modelling.leeds.ac.uk/pocketfinder/
20	Screen2	http://luna.bioc.columbia.edu/honiglab/screen2/cgi-bin/screen2.cgi
21	ConCavity	http://compbio.cs.princeton.edu/concavity/
22	ligplot	https://www.ebi.ac.uk/thornton-srv/software/LigPlus/

Table 2.

List of online protein active site predictors servers.

10. Recent advances and applications of molecular docking

Essentially, the goal of molecular docking is to use computational techniques to anticipate the structure of the ligand-receptor complex. The scientific community views this approach as a promising tool for drug discovery, especially for ligand identification, structure-based rational drug design, and prospective target predictions for many diseases. The remarkable advantages of molecular docking, specifically its capacity to predict experiments, are garnering increasing attention because of its possible uses in numerous industries [60].

10.1 In potential marine drugs discovery

Marine medications have been utilized for a long time and offer special benefits in clinical settings. Approximately 70% of the Earth’s surface is made up of oceans, where a wide variety of marine species with high activity and efficacy levels have distinctive and innovative components not found on land. Research on marine pharmaceuticals has been its specialty since 1968 when the National Cancer Institute (NCI) began to examine marine resources for potential anti-cancer activity [61]. Marine drugs developed on this basis are important in the fields of antinociceptive, anti-bacterial, cerebrovascular diseases, neurological disorders, anti-virus, malaria, anti-coagulation, cardiovascular, and anti-tumor. The abundant marine organisms can produce a variety of naturally active substances with novel structures and remarkable activities [62].

Many marine-derived active compounds have been found and investigated globally in recent decades; many of them have either been licensed for commercialization or have progressed to different phases of clinical study. In addition to having high biological activity and a variety of unique structural features, marine species’ metabolites also offer a wealth of pro-drugs and model structures for the study and creation of new medications. In recent decades, molecular docking technology a prominent technique in computer-aided drug design has been extensively utilized in the search for active ingredients and the clarification of action processes. It has also been crucial in the advancement of marine drug research and development. The in silico studies suggest that the halogenated compounds could exhibit anti-tyrosinase activity as their halogen group interacts with the catalytic hydrogen of the tyrosinase residues.

10.2 In food science

Several fundamental research about biomolecular interactions in the food matrix have gradually come to light in recent years. Since food gives the body vital nutrients including proteins, fats, carbs, and vitamins, molecular docking has a lot of applications in these fields. Furthermore, drug residues, biotoxins, and foodborne pathogens are major concerns related to food safety in food research, which is becoming more and more concerned with molecule-level analysis [16, 63, 64, 65]. The field of molecular docking has also marked its importance in discovering and identifying the natural compounds exhibiting the potential therapeutic effects or responsible for the toxic/side effects. Wang et al.’s work investigates how the binding of food items to enzymes can alter the structure and conformation of the enzyme, thereby lowering its activity [66]. Additionally, many food items compete with enzymes for substrate binding, which might lower the activity of the enzymes [67].

The docking studies investigate the antagonist/agonist or synergistic (positive or negative) reactions in the food-enzyme interaction study. The binding of lignin to α-amylase modifies the lignin structure, and the hydrogen bond identifies the hydrophobic cavity of the enzyme as the key binding site, explaining the activation of alpha-amylase. The primary binding force is the hydrophobic interaction, and the enzyme oversees the structural alterations. Understanding the relationship between an enzyme and its substrate as well as how enzyme activity is regulated is made easier by this [68]. Computational advancement also helped in exploring the bioactive peptides (BAPS) [69, 70, 71], amino acids [72, 73, 74], carbohydrates [75, 76, 77, 78], lipids [79, 80, 81], vitamins [82, 83, 84, 85], pesticides and veterinary drug residues [86, 87, 88], biotoxins [89, 90, 91, 92, 93], and foodborne pathogens [94, 95, 96, 97].

10.3 In drug repurposing approach

Drug repurposing, or the use of already-approved medications for new uses, is a useful strategy for developing novel treatments for uncommon diseases while also cutting down on the time and expense associated with drug research. Identification of possible targets is an essential first step toward the search for novel indications. Molecule docking is a popular technique for locating possible targets. It may find possible targets for a given medicine or identify potential drugs for a specific target, and it only needs structure inputs from the drug and the target.

Molecular docking is widely used in drug development and repurposing, yet there are still issues with the technique. Adding co-binders to the system, taking solvents into consideration, and optimizing the target conformation are all potential ways to increase prediction accuracy.

The study conducted by Kumar et al. in 2017 applied this approach to identify the leading antipsychotic medications bromperidol, pimozide, anisoperidone, melperone, benperidol, and anisopirol as potential candidate for drug repurposing against a variety of AD-related targets and found that the most effective potential medication for interacting with many target proteins implicated in Alzhimer’s disease is benperidol [98].

The COVID-19 pandemic has continued to bring consternation to most of the world, and the drug repurposing approach has found its application in identifying the drug against the virus. Many scientists applied the in silico approach to identify potential drugs by carrying out molecular docking on both approved drugs and substances previously tested in vivo to meet the time strain [99, 100, 101]. They have identified the molecules based on docking scores and found that the zafirlukast and simeprevir as the most promising candidates which were reported to exhibit leukotriene receptor antagonist and protease inhibitor [102].

Work done by Fortunatus et al., against schistosomiasis which is a neglected, yet common, disease of sub-tropical and tropical regions reveals that tolmetin and diflunisal are the potential drugs to be used against the disease even against the resistant species where the drugs like praziquantel or oxamniquine get failed [103]. Another study used this approach to find the best target against the Respiratory Syncytial Viral Infection, one of the most common viral infections of the respiratory airways having high mortality of younger children’s and infants. The study suggested the use of oral hypoglycemic drugs (Glipizide, Glimepiride, Glibenclamide Canagloflozin,) as potential drug candidate for the condition requiring further validation by experimental, preclinical trials, clinical trials, and route of administration optimization [104].

Mile stonework was done by Baby K. et al. in 2023, against non-small cell lung carcinomas (NSCLC), a predominant form of lung malignancy and one of the highest morbidity rates due to the deregulation of Akt, a serine/threonine kinase. A computational investigation was undertaken to identify allosteric Akt-1 inhibitors from the previously FDA-approved drugs from the databases and found that the Akt-1 allosteric site showed highest binding affinity with the dasatinib, valganciclovir, novobiocin, and indacaterol [105]. Other examples of repurposed approved drugs are shown in Table 3.

Drug	Original indication	Approved repurposed use
Allopurinol	Cancer	Gout
Aspirin	Inflammation, Pain	Antiplatelet
Amphotericin B	Antifungal	Antiparasitic
Baracitinib	Rheumatoid arthritis	COVID-19
Bromocriptine	Parkinson’s disease	Diabetes mellitus
Bupropion	Depression	Smoking cessation
Duloxetine	Depression	Stress urinary incontinence
Doxycycline	Antibiotic	Antimalarial
Eflornithine	Antitumour	Antiparasitic
Finasteride	Benign prostatic hyperplasia	Hair loss
Gabapentin	Epilepsy	Neuropathic pain
Gemcitabine	Antiviral	Cancer
Infliximab	Crohn’s Disease	Rheumatoid arthritis, Ulcerative colitis
Itraconazole	Antifungal	Anticancer
Methotrexate	Cancer	Rheumatoid arthritis
Miltefosine	Antitumour	Antiparasitic
Molnupiravir	Influenza viruses and encephalitic alphaviruses	COVID-19
Mycophenolate Mofetil	Prevention of organ transplant rejection	Lupus nephritis
Paromomycin	Antibiotic	Antiparasitic
Propranolol	Hypertension	Migraine headache
Raloxifene	Osteoporosis	Breast cancer
Remdesivir	Hepatitis–C	Ebola virus, COVID-19
Sildenafil	Angina	Erectile dysfunction
Thalidomide	Sedation, Morning sickness	Leprosy, Multiple myeloma
Tocilizumab	Rheumatoid arthritis, other autoimmune rheumatic diseases	COVID-19
Azidothymidine (Zidovudine)	Cancer	AIDS

Table 3.

Some examples of repurposed drugs [4, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116].

Developed in the 1980s, itraconazole is a triazole antifungal medication that works well against a range of systemic fungal infections. It is recognized to hinder fungal development by impairing membrane activities and inhibiting the cytochrome P450-dependent lanosterol 14-α-demethylation (14DM) in the ergosterol biosynthesis pathway. Itraconazole’s anticancer activity was initially documented by Chong et al. in 2007 because of its recently found anti-angiogenic activity.

Itraconazole had potent anticancer effects in preclinical models of medulloblastoma, basal cell carcinoma, and non-small cell lung cancer (NSCLC), either by alone or in conjunction with other anticancer medications. Itraconazole has entered multiple Phase II clinical investigations for the treatment of different forms of cancer, spurred by positive preclinical results. Positive clinical outcomes from trials for advanced lung cancer, prostate cancer, and basal cell carcinoma at Stanford University and Johns Hopkins Sidney Kimmel Comprehensive Cancer Center have been reported most recently. When compared to the control arm with pemetrexed alone, patients with progressing nonsquamous non-small-cell lung cancer who had it in conjunction with itraconazole demonstrated a significant increase in survival.

Itraconazole has demonstrated encouraging anticancer potential against several cancer types; however, its exact anticancer mechanism is yet unknown. Itraconazole has been shown to have two anticancer mechanisms thus far: it inhibits the Hedgehog signaling system in some cancer cells and angiogenesis in certain cancer cells. To better understand the specific mechanism of action of itraconazole’s anticancer effect and to aid in the development of a novel anticancer and anti-angiogenic medication, ongoing investigations are aimed at finding the molecular target of itraconazole in mammalian cells [111].

10.4 In overcoming scoring functions shortcomings through machine-learning

In silico drug discovery nowadays would not be possible without scoring functions. Still, it is a difficult challenge to accurately forecast binding affinity by scoring functions (SFs). The way scoring functions perform varies greatly depending on the target class. There is a great need for scoring systems that are based on accurate physics-based descriptors that better depict the protein-ligand recognition process. The predictive performance of classical structure-based virtual screening methodologies was either too poor or too varied, which led many practitioners to doubt its usefulness as a primary screening methodology for small compounds [117].

While widely used, classical SFs’ performance is limited because they do not capture complex non-linear relationships or take advantage of the growing amount of structural data available to improve performance [118]. To address these issues, a number of machine learning-based scoring functions (MLSFs) have been developed over the last 10 years, utilizing convolutional neural networks [119], support vector machines [120], and random forests [121]. Additionally, while classical SFs are generally linear functions with empirical, force field-based, or knowledge-based terms to estimate binding affinity [122].

Though during the past 10 years, SFs’ performance for virtual screening has improved, new approaches are required to push the envelope and make computer-driven drug discovery broadly usable. With so many experimental protein-ligand complex structures available, machine-learning scoring functions (MLSFs) appeared to have peaked in performance. Based on virtual screening benchmarks, MLSFs have shown significant performance gains over classical SFs by the prospective studies [123]. The majority of MLSFs are regressors, forecasting the receptor-ligand interaction’s numerical pKd. To lower false positive rates during virtual screening, a classification approach to affinity prediction is also a feasible option [124].

Using the Support Vector Machine for Regression (SMOReg) and Random Forest (RF) algorithms, Guedes et al. developed nonlinear scoring functions that were trained using the same physics-based descriptors chosen for the final linear scoring functions [125]. on the proteases and protein-protein interactions (PPIs). They reported the first and well-performing SVM-based scoring function specific for PPI binding sites that can be a useful tool for finding new iPPIs by enhancing the solvation and ligand torsional entropy terms, implementing in MLR DockTScore predictions, DockTScore (available at www.dockthor.lncc.br) [122].

10.5 In GPU-system accelerated molecular docking

Computational drug development is greatly aided by virtual screening, which calls for precise and effective structure-based molecular docking. Many docking techniques have been developed and tested over the past 20 years [10, 15, 17, 37], but the most of them are primarily intended for CPU-based systems; just a small number [112, 114, 115] are targeted at GPUs and heterogeneous HPC nodes [126, 127].

Fan et al. created molecular docking software building pieces and algorithms that utilize graphics processing units (GPUs) concentrating on MedusaDock [128], a versatile platform and method for protein-small molecule docking [129].

In addition to molecular docking-based HTVS, artificial intelligence (AI) methods are employed to investigate target and ligand binding potential. For instance, the MolAICal [130] program builds ligand molecules using a fragment growth technique, trains ligand fragment libraries for particular receptors using generative adversarial networks, and works with molecular docking software to determine the ideal conformation and binding energy.

Using a deep convolutional neural network technique, the DEEPScreen [131, 132].

software was trained individually for 704 targets using the SMILES of 2D molecules. Training involved using every receptor and at least 100 active ligands from the ChEMBL [133] chemical library. This resulted in the creation of unique predictive models for every target. Molecular docking-based HTVS, as opposed to AI approaches, continues to be the predominant choice in drug design due to the poor interpretability of AI methods and the limited accuracy of prediction outputs for molecules with major structural variations from the training molecules [134].

10.6 In large scale cloud docking

Molecular docking is a practical method to use structures to find ligands, and high throughput virtual (HTVS) screening is one of the basic parameters required for the approach in terms of finding a novel molecule. Large libraries are screened using this method, which quickly scores molecules for fit and ranks the database from best to worst for testing and purchase. The quantity of molecules that are commercially available has increased now by more than three orders of magnitude since 2015, opening up new possibilities for the discovery of novel chemistry and biology. After being used on more than a dozen targets, ultra-large-scale docking (LSD) has now found new compounds with activities that are frequently in the nano-molar and sporadically sub-nanomolar ranges. This has frequently resulted in molecules with intriguing in vivo activities [135].

Recent studies identifies that the docking computation cost for every target is embarrassingly high including the cost of licensed software, high GPU system and graphic system, meaning it may be distributed over several processors without significantly reducing performance [129].

Therefore, having a user-friendly and convenient online platform is crucial for regular users, particularly those who are unfamiliar with Linux operations, command lines, and other computer-based knowledge. Moreover, web services for molecular docking-based HTVS are a useful means of lowering the obstacles to HPC software utilization and accessibility while increasing productivity [136]. In addition to achieving notable gains in computational performance in recent years, public cloud computing platforms like Microsoft Azure, Google Cloud Platform, and Amazon Web Services are also anticipated to be capable of performing massively parallel computing. Bioinformatics is starting to use the cloud as it allows users to employ thousands of CPU cores and GPU accelerators arbitrarily, and because cloud images make it very easy to use different software kinds. As an example of an HPC cloud environment, Ohue et al. transferred the original protein-protein interaction prediction (protein–protein docking) software, MEGADOCK, onto Microsoft Azure and found it more efficient and cost effective [137].

Zhixiong et al. also indulged the cloud computing in marking the use of free energy perturbation (FEP) for binding affinity prediction between candidate compounds and their biological targets which has grown significantly. FEP applications do have certain drawbacks, though, such as their high cost, lengthy wait times, restricted scalability, and narrow range of application situations. Using enhanced simulation protocols, they have created XFEP, a scalable cloud computing platform for relative and absolute free energy forecasts, to address these issues. With the help of XFEP, large-scale FEP computations may be carried out more effectively, scalable, and affordably. For instance, 5000 compounds can be evaluated in a week utilizing 50–100 GPUs, with a computing cost that is nearly equal to the cost of synthesizing just one novel chemical.

Further, these abilities can be combined with artificial intelligence methods for goal-directed molecule generation and assessment to explore new avenues for applications in hit identification, hit-to-lead, and lead optimization drug discovery stages. These stages involve not only structure exploitation within the given chemical series but also evaluation and comparison of entirely unrelated molecules during structure exploration in a larger chemical space [138].

11. Conclusion

Making helpful predictions utilizing various tools that could be obtained through molecular modeling using computational techniques is one of the primary goals of computational medicinal chemistry. This fact is crucial when putting strategies into practice for drug development projects or protocols because of the various jobs and difficulties involved in finding the right hit compound. In this field, molecular docking plays a significant role and consistently offers benefits. Advances in computational power combined with imaging methods, artificial intelligence strategies, and protein fold prediction tools (e.g., AlphaFold), suggest a golden age for the application of the computational approaches to drug discovery, wherein knowledge of the function and shape of proteins within cells provides better protein structures for molecular docking, contributing to the development of targeted and selective new therapeutics while also lowering the expenses of testing.

Comparing repurposing molecules to traditional drug research and development approaches reveals certain advantages like Hasten’s drug development, reduction in costs and risks, and personalized medicine while it still lacks the identification of the correct dose, potential pharmaceutical and toxicological issues, and the patenting rights. The importance of achieving notable gains in computational performance in recent years, through application of machine learning, by GPU system, overcoming the scoring function limitations, and public cloud computing platforms like Microsoft Azure, Google Cloud Platform, and Amazon Web Services in running large-scale molecular docking screens is now pragmatic and does not require any commercial software exploring new avenues for applications in hit identification, hit-to-lead, and lead optimization drug discovery stages.

Conflict of interest

“The authors declare no conflict of interest.”

Glossary

FDA	Food and Drug Administration
COVID-19	coronavirus disease 2019
HTS	high-throughput screening
NCBI	National Center for Biotechnology Information
GA	genetic algorithm
ADAM	advanced design and modeling
FRED	fast rigid exhaustive docking
GLIDE	grid-based ligand docking from energetics
SLIDE	screening for ligands by induced-fit docking, efficiently
GOLD	genetic optimization for ligand docking
MOE	molecular operating environment
PLANTS	protein-ligand ANT system
QM	quantum mechanics
SDF	structured data file
PDBQT	protein data bank, partial charge, and atom type partial charges (q) and atom types (t)
XYZ Coordinates	Cartesian coordinates
DNA	deoxyribonucleic acid
RNA	ribonucleic acid
BAPS	bioactive peptides
AD	Alzheimer’s disease
NSCLC	non-small cell lung carcinomas
Drug repurposing or drug repositioning or drug reprofiling	the technique of using an existing approved drug candidate for a new treatment or medical indication for which it was not indicated before
Serendipity	accidental, unexpected, and fortunate discoveries
Off-label use	drug used for different purposes than those for which they were initially licensed
Molecular docking	interaction of two or more molecules to give the stable adduct
De Novo drug development	methodology that creates novel chemical entities based only on the information regarding a biological target (receptor) or its known active binders
Drug design	the approach of finding drugs by design, based on their biological targets
Drug discovery	process of identifying chemical entities that have the potential to become therapeutic agents