Computer-Aided Drug Design and Development: An Integrated Approach

Abstract

Drug discovery and development is a very time- and resource-consuming process. Comprehensive knowledge of chemistry has been integrated with information technology to streamline drug discovery, design, development, and optimization. Computer-aided drug design is being utilized to expedite and facilitate hit identification, hit-to-lead selection, and optimize the absorption, distribution, metabolism, excretion, and toxicity profile. Regulatory organizations and the pharmaceutical industry are continuously involved in the development of computational techniques that will improve the effectiveness and efficiency of the drug discovery process while decreasing the use of animals, cost, and time and increasing predictability. The present chapter will provide an overview of computational tools, such as structure-based and receptor-based drug designing, and how the coupling of these tools with a rational drug design process has led to the discovery of small molecules as therapeutic agents for numerous human disease conditions duly approved by the Food and Drug Administration. It is expected that the power of CADD will grow as the technology continues to evolve.

1. Introduction

“This is a fantastic time to be doing drug discovery. We have an incredible wealth of knowledge that has been generated over the past few years.”

Drug discovery and development (DD&D) is a lengthy and complex process that takes around 12–15 years and costs up to multi-million dollars for a drug to reach the market. Interdisciplinary DD&D begins with the identification and validation of a suitable drug target, followed by a hit to lead discovery and optimization, and finally preclinical and clinical studies. Despite the huge investments and time incurred for the discovery of new drugs, the success rates are too low that only five out of 10,000 compounds make their way to reach human testing after preliminary evaluation in animals and only one of five compounds reaches final clinical studies. Further, a majority (40–60%) of the drug failure has been observed at a later stage of the DD&D process due to a lack of optimal pharmacokinetic properties, that is, absorption, distribution, metabolism, excretion, and toxicity (ADME/Tox). This all suggested and urged the need to develop new methodologies to facilitate and expedite the DD&D process [1].

Advances in computational techniques and parallel hardware support have enabled computer-aided drug design (in silico) methods, that are being used by leading pharmaceutical companies and research groups to speed up the DD&D process. It entails:

• Usage of computing power to simplify drug discovery and development process

• Leveraging the chemical and biological information about targets and/or ligands to identify and optimize new drugs

• Designing in silico filters to remove compounds with poor activity and/or poor ADMET

• Selection of the most promising candidates.

Broadly classified computational methods include combinatorial chemistry; high throughput screening; virtual screening; pharmacophore modeling structure-based drug design (drug-target docking); ligand-based drug design (pharmacophore, a 3-D spatial arrangement of chemical features essential for biological activity)—quantitative structure-activity and structure-property relationships; quantum mechanics and in silico evaluation methods [2]. The commonly and widely used CADD approaches are structure-based and ligand-based drug design approaches, to identify suitable lead molecules in the drug discovery process. The structure-based drug design (SBDD) relies on the three-dimensional (3D) structure of the target receptor and its active sites to understand the molecular interaction between the receptor and ligand. While the ligand-based-drug design (LBDD) depends on the knowledge of ligands interacting with the given target receptor. A few marginally active or better compounds may be found, and then chemical similarity searching techniques are used to find more compounds that can be assayed. Finding some of the more active compounds further computationally techniques are applied to identify more potent compounds with favorable ADME/T [2, 3].

The application of rational drug design as an integral part of CADD provides useful insights into the understanding of the binding affinity and molecular interaction between target and ligand. Additionally, lead identification in pharmaceutical research has been facilitated by the availability of the super-computing facility, parallel processing, and advanced programs, algorithms, and tools. Furthermore, recent advancements in artificial intelligence (AI) and machine learning methods have greatly aided in analyzing, learning, and explaining the pharmaceutical-related big data in the drug discovery process [4].

2. Structure-based drug design

2.1 Docking

Understanding the principles by which small-molecule (ligands) recognize and interact with a receptor (macromolecules target) is of great importance in the pharmaceutical research and development (R&D) process. The availability and systematic use of the three-dimensional (3D) structure of the therapeutic target proteins and exploration of the binding site cavity form the basis of structure-based drug design (SBDD). This specific and fast approach identifies the lead molecules followed by their optimization and helps to understand disease at a molecular level. Some of the common methods employed in SBDD include structure-based virtual screening (SBVS), molecular docking, and molecular dynamics (MD) simulations. In case of the non-availability of structural information of the desired target, another computational approach, that is, homology modeling can be utilized to build the model of the needed target [5].

Molecular docking is one of the most commonly used techniques in SBDD, because of its ability to predict with a significant degree of accuracy the conformation of small-molecule ligands within the applicable target binding site. Two approaches are primarily used to perform molecular docking, the simulation approach, and the shape complimentary approach. The first approach utilizes computer simulations wherein the energy profiling is assessed for ligand-target docked conformer. Whereas, the second approach employs the practice that calculates surface complementarity between ligand and target. The brief and main properties of both approaches are described in Table 1 [6].

Simulation approach	Shape complementarity approach
Interaction energy as per ligand-receptor pair is calculated.	Estimation of complementarity between ligand and receptor surface.
Ligand is allowed to fit into the receptor’s groove based upon minimum energy consideration.	Solvent accessible topographic features of ligand and receptor in terms of matching surface are described.
Every move of ligand into receptor’s pocket for best fitting generates energy as Total Energy of System and is compared to find out best-docked conformer with minimum energy.	It involves the surface representation of receptor and ligand (i.e., surface construction and smoothing), features/curvature calculation followed by docking and scoring contingent on geometric complementary criteria.
More compatible to accept ligand flexibility in the molecular modeling tool.	Both types of docking; flexible docking and rigid docking are feasible.
Requires much longer time as large energy profiling needs to be estimated.	Rapid scanning of a large number of ligands for the binding on its target in a few seconds and hence provides quick and robust outcomes.

Table 1.

Molecular docking approaches.

2.2 Methodology

It is a cyclic process with basic steps of the preparation of target structure, binding site prediction, ligand preparation using generated library of synthesized compounds, molecular docking, binding free energy calculations, analysis based on these scoring functions, and molecular dynamic simulation. In the last few decades, advancement in structural elucidation techniques, such as X-ray and NMR, has increased the availability of protein structure deposits in protein data bank (PDB). However, some of the target protein structures have not been solved to date and computational techniques, such as comparative homology modeling, ab initio modeling and threading are successful in interpreting the structures of such proteins from their sequences. After the target selection, it’s important to identify the ligand-binding site, a prerequisite step for carrying out specific docking. The information on the binding sites can be obtained from the X-ray crystallographic structures of proteins co-crystallized with substrates or inhibitors or through site-directed mutagenesis study. In the absence of the experimental information about the binding site of many proteins, plenty of software and webservers such DoGSite Scorer, CASTp, DEPTH, and NSiteMatch, allows us to predict the putative binding sites of the isolated and purified protein [7]. For a particular selected prepared protein, a library of the compounds retrieved from chemical databases, ZINC database, DrugBank, PubChem, or synthesized molecules in the research laboratories can be tested. Also, it preferred to perform the docking on drug-like compounds being filtered using Lipinski’s rule of five and ADMET (absorption, distribution, metabolism, excretion, and toxicity) parameters, including risk parameters, such as acute rat toxicity, carcinogenicity, hepatotoxicity, and mutagenicity [8].

2.3 Interpretation and correlation

The ultimate purpose of the SBDD is to visualize ligands with specific electrostatic (charge distribution) and stereochemical (bulky or small) attributes to achieve high receptor-binding affinity. First, developed algorithms for molecular docking in 1980, enabled us to understand crucial molecular events, such as ligand-binding modes and their corresponding intermolecular interactions that stabilize the ligand-receptor complex [9]. The availability of 3D protein (macromolecular) structures further facilitated a careful inspection of the binding site topology, as well as the presence of cavities, clefts, and sub-pockets. Investigations arrange and rank the docked compounds based on the specific scoring functions of ligand-receptor complexes after quantitative predictions of their binding energetics [9]. Molecular docking programs identify the most likely binding conformations after cyclical steps of exploring the large conformational space representing various potential binding modes and accurate prediction of the interaction energy coupled with each of these predicted binding conformations [10].

These molecular modeling procedures are followed by the synthesis of the most promising compounds, and their biological properties evaluations using diverse in vitro and in vivo experimental protocols. Once the active compounds have been identified, the 3D structures of such ligand-receptor complexes from docking studies allow the observation of numerous intermolecular features supporting the process of molecular recognition. Structural descriptions of these complexes are helpful in investigating the binding conformations, key intermolecular interactions, characterization of unknown binding sites, mechanistic studies, and the elucidation of ligand-induced conformational changes [11]. Then biological activity data can be justified and correlated to structural information once a ligand-receptor complex has been determined.

In this way, the SBDD process in CADD starts over with new steps to design ligand by incorporating necessary molecular modifications for efficient or increased affinity toward the binding site. And selective adjustments of a validated drug target by high-affinity ligands ultimately lead to the desired pharmacological and therapeutic effects by interfering with specific cellular processes. The flexibility of the target receptor always remains an essential aspect to be considered throughout the modeling phase, as substantial conformational changes can occur upon its binding with the ligand. To address the issue, two modified approaches flexible docking (flexible-ligand and flexible-protein search docking) and molecular dynamics were further tried and explored. Stochastic method, systematic method, and simulation method are the three commonly used algorithms for ligand flexibility in the case of the flexible-ligand search docking method, whereas flexible-protein docking usually depends on molecular dynamic (MD) and Monte Carlo (MC) methods [12].

2.4 Application and tools

SBDD as a computational technique has greatly helped many pharmaceutical industries and medicinal chemists in the discovery of several drugs available on the market. Few of them include the discovery of amprenavir as a potential inhibitor of the human immune deficiency virus (HIV) protease using homology protein modeling and MD simulations; norfloxacin a commonly used antibiotic against urinary tract infection using SBVS, isoniazid (antituberculosis drug) an enoyl-acyl-ACP reductase (InhA) inhibitor discovered through SBVS and pharmacophore modeling, and flurbiprofen, a nonsteroidal anti-inflammatory (NSAID) drug used against rheumatoid and osteoarthritis targeting cyclooxygenase-2 (COX-2) discovered through molecular docking approach, etc. [13]. Some of the freely available software tools for docking study are enlisted in Table 2 summarizing their algorithms, scoring functions, and advantages.

S. No.	Software tools	Algorithm	Scoring term	Advantages
1.	Glide	Monte Carlo	Glide score	Lead discovery and lead optimization
2.	AutoDock	Lamarckian genetic algorithm	Empirical free energy function	Adaptability to user-defined input
3.	GOLD	Genetic algorithm	GoldScore, ChemScore, ASP CHEMPLP	Allows atomic overlapping between protein and ligand
4.	ICM	Monte Carlo minimization	Virtual library screening scoring function	Allows side-chain flexibility to find a parallel arrangement of two rigid helixes

Table 2.

List of software tools for docking and their algorithms [14].

3. Ligand-based drug design

Ligand-based drug design is an extensively used approach in computer-aided drug designing, especially when the three-dimensional structure of the target receptor is not available. The information derived from a set of active compounds against a particular target receptor can be used in the identification of structural and physicochemical properties accountable for the given biological activity, based on the fact that structural resemblances correspond to similar biological functions. Pharmacophore modeling [15] and quantitative structure-activity relationships (QSARs) [16] are some of the conventional techniques used in ligand-based virtual screening (LBVS).

Generated pharmacophore model elucidates the spatial arrangement of chemical features in ligands that are essential for interaction with the target receptor. H-bond donors/acceptors, hydrophobic areas, aromatic ring systems, and +ve/−ve charged ionizable groups are some of the chemical features used in pharmacophore modeling. Ligands with different scaffolds but the similar spatial arrangement of the above-mentioned key interacting functional moieties can be identified using pharmacophore-based virtual screening [15]. The conformation of the active molecules within the target binding site can be integrated into the pharmacophore model for further application in QSAR studies in the molecular alignment stage.

3.1 Quantitative structure-activity relationships (QSARs)

Quantitative structure-activity relationship (QSAR) analysis is the most commonly utilized LBDD approach and correlates the variations in the bioactivity of the compounds with the changes in molecular structure. A statistically significant model is being constructed using these correlation studies, and the final model can be utilized to predict the biological activity of new molecules [17]. They are widely used in the drug discovery process in the hit to lead identification or lead optimization.

The technique was developed more than 50 years ago by Hansch and Fujita (1964), where affinities of ligands to their binding sites, rate, and inhibition constants were correlated with other biological endpoints with the atomic, group, or molecular properties, such as lipophilicity, electronic parameters, polarizability, and steric properties (Hansch analysis), or with some structural features (Free-Wilson analysis) [18]. But, the limitation of the classical approach in designing a new molecule on account of the lack of understanding of the 3D structure of the molecules urged the scientists to look for an alternative. So, an extension to the existing classical Hansch and Free-Wilson approaches emerged as 3D-QSAR, which exploits the three-dimensional properties of the ligands to predict their biological activities. Since then and until now, QSAR continues an efficient approach for building mathematical models to find a statistically significant correlation between the chemical structure and continuous (pIC₅₀, pEC₅₀, Ki, etc.) or categorical/binary (active, inactive, toxic, nontoxic, etc.) biological/toxicological property using regression and classification techniques, respectively [19].

In the last decades, QSAR has undergone several transformations, ranging from the dimensionality of the molecular descriptors (Table 3) and different methods (Table 4) for finding a correlation between the chemical structures and the biological property. Based on how the descriptors are derived, QSAR can be classified into six different types (Table 5) [20].

Hydrophobic descriptors	Partition coefficient, Hansch’s substitution constant, distribution coefficient, hydrophobic fragmental constant, solubility parameter
Electronic descriptors	Hammett constant, Taft’s inductive (polar) constant, ionization constant, Swain and Lupton field parameter
Steric descriptors	Taft’s steric parameter, Vander Walls volume, molar volume, molar refractivity, and Vander Walls radius.
Spatial descriptors	Shadow indices, principle moment of inertia, the radius of gyration
Quantum chemical descriptors	Atomic net charge, super delocalizability, energy of lowest unoccupied molecular orbital. The energy of the highest occupied molecular orbital

Table 3.

Types of descriptors in different QSAR.

Linear method	Linear regression (LR), multiple linear regression (MLR), partial least squares (PLS), principal component analysis (PCA), and principal component regression (PCR).
Nonlinear	k-nearest neighbors (kNN), artificial neural networks (ANN), and Bayesian neural nets [20].

Table 4.

QSAR classification based on the methodology.

One dimensional (1D)	Correlates the activity with global molecular properties, such as pKa and log P
Two (2) dimensional	Correlates the activity with structural patterns, such as connectivity indices, and 2D pharmacophore without considering the 3D representation of these properties
Three (3) dimensional	Activity is correlated with noncovalent interaction fields surrounding the molecules
Four (4) dimensional	It further includes set of ligand configurations in 3D QSAR.
Five (5) dimensional	Explicitly represents the different induced fit models in 4D QSAR.
Six(6) dimensional 6D	Additionally, it incorporates the solvation models in 5D QSAR

Table 5.

QSAR classification based on parameter correlated.

Depending on the quantity of dataset, QSARs can be classified and designed as local or global. A local QSAR is trained on a small and congeneric series of chemical structures, whereas a large and structurally diverse set of chemicals are used for global QSAR [21]. The development of the QSAR model follows a general workflow as shown in Figure 1, starting from the collection of the dataset, its curation, and preparation, followed by the generation and selection of chemical descriptors to be used as independent variables. Collected data is divided into training and test sets, and the QSAR model is built using a training set consisting of chemical structures and their related experimental data. The chemical structures are represented by chemical descriptors, which are used as independent variables in the model. Statistically significant model (s) are built within the defined applicability domains (AD) based on field-based or atom-based approaches. In field QSAR, the molecule atoms are treated as fields and classified as either H-bond acceptor, H-bond donor, electrostatic or stearic. On the other hand, in atom-based QSAR, molecule atoms are considered spheres and can predict the results for H-bond donor/acceptor, non-polar/hydrophobic, negative/positive ionic, or electron-withdrawing groups. Both methods can predict the change in activity by the fields or spheres in molecules with various types of substitution [22].

Figure 1.

Workflow of ligand based drug designing.

Validation is the process to access the reliability and relevance of a particular approach, method, or process established for a defined purpose. Certainly, it is possible to calculate a large number of descriptors using various software tools, however, one cannot ignore the risk of chance correlations with the higher number of variables incorporated in the final model as compared to the number of compounds for which the model has been constructed. And with diverse optimization procedures, one can try to get models that can fit the experimental data well, but there is always a chance of overfitting. Thus, all indicate the necessity to validate the developed models for their robustness and predictivity. And QSAR models are fundamentally judged for their predictivity, which represents how well they are capable to forecast the end-point values of molecules being not employed to develop the correlation. QSAR models can be validated using two major strategies: (i) internal validation based on the training set molecules, and (ii) external validation using the test set compounds after splitting the whole data set into training and test sets. Validation and analysis are characterized in terms of statistical parameters, such as regression values (R2), variance ratios (F), standard deviations (SD), root mean square error of the test set (RMSE), anticipated activities of the test set (Q2), and Pearson-R, were used to validate and analyze the model generated [18].

3.2 Tools

Quantitative structure-activity relationships (QSAR) have been applied for decades in the development of new drugs. Although a QSAR does not completely eliminate the trial-and-error factor involved in the development of a new drug, it certainly decreases the number of compounds synthesized by facilitating the selection of the most promising examples. A number of successful QSAR studies have been carried out using various platforms and helped in building models for predicting chemical, biological and toxicological activities. Some of these are listed in Table 6.

Software	Description	Availability
ACD/Tox Suite	Prediction of toxicity	Commercial, free web service
ADMET Predictor	ADMET properties prediction	Commercial
AZ Orange Machine learning platform for	QSAR modeling	Free
3D-QSAR	3D QSAR models	Free
BioTriangle	Web-based platform to calculate the molecular descriptors	Free
BioPPsy	Prediction of pharmacokinetic properties of drug candidates by QSPR modeling	Free
BlueDesc	Molecular Descriptor Calculator	Free
CAESAR	Developments of toxicity models	Free
CACTVS	Molecular descriptor calculator	Free
CODESSA	Generate predictive QSAR models from quantum chemical, topological and electrostatic descriptors	Commercial
ChemDes	Calculations of descriptor and fingerprint (web-based)	Free

Table 6.

Software’s and their features [14].

4. In silico 5-AlPHA reductase activity prediction using SBDD

Benign prostatic hyperplasia (BPH), a common condition of males over the age of 50 is characterized by enlargement of the prostate and results in urinary obstructions. Dihydrotestosterone (DHT), the 5α-reduced metabolite of testosterone (T) (Figure 2) has been implicated as a causative factor in this progression of BPH. 5-alpha reductase (5AR) a membrane-bound NADPH-dependent enzyme is responsible for the irreversible conversion of androgen T into DHT in the prostate. And inhibition of 5AR represents a logical treatment for controlling the BPH by diminishing the concentration of DHT in the prostate and is expected to improve the pathology of this disease [23].

Figure 2.

Action of 5-alpha reductase (5AR).

In the past few decades, numerous nonsteroidal and steroidal compounds have been prepared as competitive or noncompetitive inhibitors of 5AR [24, 25]. In 1992, Finasteride (MK-906) was approved as the first 5-alpha reductase inhibitor (5ARI) in the United States for the clinical management of BPH. It is a competitive inhibitor of 5AR type-2 and at the clinical doses of 5 mg/day, it forms a stable complex with a 10-fold higher affinity than type 1 and decreases the prostatic DHT level and prostate volume by 70–90%, in human beings. Unlike finasteride, another molecule, that is, dutasteride (Figure 3) a nonselective competitive inhibitor of both isozymes. 5AR type-1 and type-2 were approved in 2002 by the USA for the symptomatic treatment of BPH [26].

Figure 3.

Clinically approved 5-alpha reductase inhibitors.

Since then, only finasteride and dutasteride are being used, but their long-term treatment has been found to be associated with the development of decreased libido, impotence or ejaculatory dysfunction, or, breast enlargement while rashes, insulin resistance, type 2 diabetes, kidney dysfunction, and other metabolic dysfunctions [26]. Thus, the identification of highly efficacious and selective inhibitors of 5AR for use in the treatment of BPH has engendered considerable interest from research groups in several laboratories. Attempts have been made in our laboratory to synthesize and characterize novel 5ARIs (ND-1 to ND-6) (Figure 4).

Figure 4.

Novel 5-alpha reductase inhibitors synthesized in our laboratory.

Structure-based drug designing approach, that is, docking (in silico) was performed on the newly synthesized compounds (ND-1 to ND-6) against PDBID: 4AT0 of 5AR receptor, using extra precision GRIP docking feature in BIOPREDICTA module available in the molecular design suite of Vlife MDS software package, version 4.6. Reference drug finasteride (FN) known to possess an affinity for the 5AR receptor was also included in the docking studies for comparing the docking results. Studies have shown that all the synthesized compounds have been found to bound better with the 5AR receptor affording dock score from −59.75 to −74.91 than reference, finasteride −57.09 (Table 7).

S. No.	Compounds	D-score	No. of residues	Hydrophobic	Aromatic	Hydrogen
1	ND-1	−66.23	5	ALA60A, ALA59A, ILE31A, SER471A, ALA32A	—	—
2	ND-2	−62.87	6	ALA59A, ALA292A, THR175A, ALA264A, THRS60A, LEU472A	—	—
3	ND-3	−74.91	7	ALA59A, ALA63A, THRS175A, THRS60A	—	GLY241A, GLY245A
4	ND-4	−60.68	8	ALA63A, ALA59A, THRS175A, ILE31A, SER471A, ALA32A, GLY30A	—	GLY310A
5	ND-5	−59.75	8	ALA63A, ALA59A, ALA264A, ALA45A, LEU472A, GLY475A, ALA32A, GLY455A	—	—
6	ND-6	−62.79	6	ALA59A, THR175A, ALA63A, SER471A, ALA264A	HIS268A	—
7	FN	−57.09	10	GLY176A, LEU62A, ALA59A, SER471A, ILE31A, LEU472A, GLY475A, GLY455A, ARG456A, THR175A	—	—

Table 7.

Docking score of novel 5ARIs.

The docking score has been observed in order ND3 > ND1 > ND2 > ND6 > ND4 and ND5. Further, similar docking behavior has been observed among all the synthesized compounds in comparison to FN, by interacting hydrophobically with common amino acid residues THR175A, ALA59A, ALA63A, SER471A, and ALA264A. Additional aromatic interactions have also been observed only in ND-6 for amino acid HIS268A. Among all the newly synthesized derivatives ND-1 to ND-6, ND-3 has been found to bound best with the 5AR receptor affording the highest D-score of −74.91 than FN. This high score of −74.91 can be attributed to its strong hydrogen bonds between NO₂ at the p-position of the benzene ring with amino acid residue GLY241A and GLY245A. The binding pose of the fitted ligands was visualized extending deep into the active site pocket and are presented in the Figure 5a–f.

Figure 5.

(a, c, and e) Grip docked pose of standard drug FN, ND-1, and ND 3 into the active binding site of 5AR receptor; (b, d, and f) 2D representation of the FN, ND-1, and ND 3 with active amino residues of 5AR receptor.

5. Conclusion

In the pharmaceutical industries, Research and Development is undergoing a lot of technological changes, and there is pressure to make the investment pay off. There is a massive demand to sensibly use the big amount of chemical and biological-related data produced in the process. Careful use of chemoinformatics techniques and software is becoming crucial in drug discovery success and an evolving field with many facets. In the past few years, it has led to the discovery of small-molecule therapeutic agents with activity directed against target proteins critical in the presentation of numerous disease conditions and/or have supported their clinical evaluation. Coupling more sophisticated computer software and hardware technologies with a rational drug design process has become an indispensable tool for the development of effective therapies and it yields information that is not easy to obtain in laboratory analysis, and, furthermore, is typically (much) less costly, save time, money, and resources. It is expected that the power of CADD will grow as the technology continues to evolve.