Advancements in small molecule drug design: A structural perspective

Abstract

In this review, we outline recent advancements in small molecule drug design from a structural perspective. We compare protein structure prediction methods and explore the role of the ligand binding pocket in structure-based drug design. We examine various structural features used to optimize drug candidates, including functional groups, stereochemistry, and molecular weight. Computational tools such as molecular docking and virtual screening are discussed for predicting and optimizing drug candidate structures. We present examples of drug candidates designed based on their molecular structure and discuss future directions in the field. By effectively integrating structural information with other valuable data sources, we can improve the drug discovery process, leading to the identification of novel therapeutics with improved efficacy, specificity, and safety profiles.

Introduction

The field of small molecule drug design is a dynamic and rapidly evolving discipline, with advancements in computational tools and methodologies significantly enhancing our ability to design and optimize small molecule drug candidates. This review provides a unique and comprehensive perspective on the current state-of-the-art small molecule drug design, with a particular focus on the role of molecular structure in determining the pharmacological properties of small molecule drugs. While the value of small molecule development is well known, our review highlights the novel approaches and insights that have emerged in recent years.

We delve into the importance of accurate three-dimensional (3D) protein structure prediction in small molecule drug discovery, comparing two distinct prediction methods: homology modeling and de novo modeling. We also highlight the critical role of the ligand binding pocket in the structure-based design of small molecule drugs. Furthermore, we explore the various structural features and properties that are commonly employed in the design and optimization of small molecule drug candidates, such as functional groups, stereochemistry, and molecular weight. We also discuss the computational tools and methods used to predict and optimize these structural properties, including molecular docking and structure-based virtual screening. In addition, we present recent examples of small molecule drug candidates that were designed based on their molecular structure, discussing their chemical structure, mechanism of action, and pharmacological properties. We also touch upon emerging trends and future directions in small molecule drug design, such as the advent of new computational tools, the identification of emerging targets, and the exploration of novel therapeutic applications.

This review is distinct in its approach to small molecule drug design, emphasizing the integration of structural information with other tools and data sources to achieve more efficient and effective small molecule drug discovery and development. In essence, this review offers a fresh perspective on small molecule drug design, underscoring the importance of molecular structure in drug development and highlighting the latest advancements and future directions in this exciting field.

Prediction of 3D protein structures in small molecule drug discovery

The determination of protein structure can be achieved through various methods such as X-ray diffraction of protein crystals, cryo-electron microscopy, nuclear magnetic resonance (NMR), and prediction techniques. The accurate prediction of 3D protein structures has become a pivotal aspect in the realm of drug design, garnering significant attention in recent years.¹ Understanding the 3D structure of a protein is of paramount importance in elucidating the mechanisms underlying drug binding and facilitating the design of compounds with enhanced specificity and potency. Notably, the field of drug discovery has undergone a revolution due to remarkable advancements in computational techniques, facilitating the prediction of protein structures with exceptional precision.

One of the most widely used computational methods for predicting protein structures is homology modeling.² Homology modeling, also known as comparative modeling, utilizes the known structures of related proteins to predict the structure of the target protein.³ This method assumes that evolutionarily related proteins have similar structures and that the structure of a protein is more conserved than its sequence.⁴ The process of homology modeling involves several steps including template selection, sequence alignment, model building, and model refinement.⁴ Each of these steps is essential for generating an accurate 3D model of a protein from its amino acid sequence (Figure 1).

FIGURE 1. Schematic diagram of the computational methods for predicting 3D protein structures in small molecule drug discovery.

Homology modeling utilizes known structures of related proteins to predict the structure of the target protein. The process involves template selection, sequence alignment, model building, and model refinement. De novo protein structure prediction, on the other hand, involves predicting the structure of a protein from scratch without relying on known structures. The process involves fragment assembly, model building, and model refinement. Both methods are critical for understanding the mechanism of drug binding and designing drugs with improved specificity and potency.

1. Template selection: The first step is to select an appropriate template structure from the protein structure database. The template structure should have a high degree of homology with the target protein sequence, and should have a similar function or fold. Several methods, such as FASTA and BLAST, are used to identify homologous sequences in the protein structure database.⁵

2. Sequence alignment: The second step is to align the target protein sequence with the selected template structure. The alignment should be optimized to ensure that the conserved regions between the target and template structures are aligned correctly. Several algorithms, such as ClustalW and MUSCLE, are used to perform sequence alignment.⁶

3. Model building: In this step, a 3D model of the target protein is generated based on the aligned target and template sequences. The model is built by superimposing the target sequence onto the template structure and adjusting the side chains to fit the target sequence. Several software programs, such as MODELLER and SWISS-MODEL, are used for model building.⁷

4. Model refinement: Once the initial model is generated, it is refined to improve its accuracy. This step involves optimizing the side-chain orientations, minimizing the energy of the structure, and performing structural validation to ensure that the model is biologically plausible. Several software programs, such as CHARMM and GROMACS, are used for model refinement.^8,9

Another method is de novo protein structure prediction (Figure 1), which involves predicting the structure of a protein from scratch without relying on known structures.¹⁰ This method is based on physical principles, such as energy minimization and free energy calculations, to predict the structure of a protein.^11,12 De novo protein structure prediction is the algorithmic process of predicting the tertiary structure of a protein from its primary sequence of amino acids, without the use of templates from previously solved structures.¹³ The process involves several steps including fragment assembly, model building, and model refinement.¹⁴ This process can be guided by knowledge-based energy functions or machine learning (ML) algorithms.^14,15

The first step in de novo protein structure prediction is fragment assembly. This involves breaking down the protein sequence into smaller fragments and predicting the relative orientations of these fragments in 3D space. Fragments can be assembled using either physics-based methods or statistical methods. Physics-based methods calculate the energy of the system to determine the optimal fragment orientation, whereas statistical methods use the likelihood of a particular fragment assembly based on known structures.^16,17 Once fragments are assembled, the next step is model building, where a complete 3D model is constructed by joining the fragments together in a way that satisfies spatial constraints such as steric clashes and bond angles. The model building process can be guided by a knowledge-based energy function, which provides a score for the quality of the model, or by ML algorithms that learn from previous successful predictions.^14,18 The final step in de novo protein structure prediction is model refinement, which involves optimizing the 3D model to improve its accuracy and overall quality. This can be achieved through molecular dynamics (MD) simulations, energy minimization algorithms, or ML methods.^19,20

Homology modeling is generally more accurate than de novo protein structure prediction, as it relies on the availability of related protein structures.² Homology modeling is considered to be the most accurate method for protein structure prediction, and is often used in drug design for screening of large libraries.² Homology models contain sufficient information about the spatial arrangement of important residues in the protein, and accurate predictions can be obtained if the template and query sequences have high sequence similarity.²¹ However, homology modeling has limitations, especially for proteins with only remote homologs or for proteins with no known structural homologs.²² Accurate template-query alignment and template selection are still very challenging for these proteins, and it can be difficult to obtain accurate models. In such cases, de novo modeling can be useful as it does not rely on the availability of related protein structures.²⁰ De novo modeling can also be used in cases where the available templates do not provide a good match for the query sequence or where the available templates have low sequence identity to the query.²⁰ De novo modeling involves predicting the protein structure from scratch, and it can be challenging due to the complexity of the protein folding problem. However, recent advances in deep learning-based prediction have shown that more accurate models can be generated by extending deep learning-based prediction to inter-residue orientations in addition to distances, and the development of a constrained optimization by Rosetta.¹²

In conclusion, accurate protein structure prediction is crucial for designing small molecule drug candidates that bind specifically to the target protein, without causing off-target effects.²³ The use of computational methods such as homology modeling and de novo structure prediction has enabled the prediction of protein structures with high accuracy, which has revolutionized the field of drug discovery. The ability to predict protein structures accurately has also led to the development of new drugs targeting proteins that were previously considered ‘undruggable’.²⁴

The crucial role of ligand binding pocket in the structure-based design of small molecule drugs

The ligand binding pocket plays a crucial role in the structure-based design of small molecule drugs. This design approach starts with the assumption that a drug molecule exerts its biological activity through specific binding to a macromolecular target receptor, typically a protein. The binding pocket refers to a cavity or depression on the surface of the target protein where the ligand binds and interacts with the protein. The structural and chemical complementarity between the ligand and the receptor within the binding pocket is a prerequisite for strong and selective binding. By fitting precisely into the binding pocket, the ligand modulates the function of the protein, potentially leading to therapeutic effects. Therefore, understanding and characterizing the binding pocket is crucial for designing small molecule drugs.

One of the primary goals in structure-based drug design is to identify ligands that can bind with high affinity and specificity to the target protein’s binding pocket. This process involves several steps including target selection, molecular docking, and virtual screening. Molecular docking simulations use computational methods to predict the binding modes and binding affinities of small molecules within the binding pocket of the target protein. These simulations help identify potential drug candidates and optimize their interactions with the binding pocket. Advancements in structural biology techniques, such as X-ray crystallography and cryo-electron microscopy, have facilitated the determination of protein–ligand complex structures, providing valuable insights into the binding interactions.²⁵ These structural insights allow medicinal chemists to design and optimize small molecules that can more effectively fit and interact with the binding pocket. Additionally, the druggability of the binding pocket is a crucial consideration in small molecule drug design. The druggability refers to the likelihood of successfully developing a drug that can bind to and modulate the target protein’s function. Factors such as the size, shape, and physicochemical properties of the binding pocket influence its druggability.²⁵ Computational methods, including MD simulations and virtual screening, can aid in assessing the druggability of the binding pocket and guiding the design of small molecules with optimal properties for binding.

There are several key points highlighting the importance of the ligand binding pocket in structure-based drug design.

1. Specificity and affinity: The ligand binding pocket provides a 3D environment that accommodates the ligand with high specificity and affinity. The shape, electrostatic properties, and chemical composition of the pocket determine the ligand’s interactions and binding strength with the protein.

2. Rational ligand design: Knowledge of the ligand binding pocket’s structure allows for rational ligand design. By studying the pocket’s characteristics, such as its size, shape, and residues lining the pocket, researchers can design small molecules that fit optimally within the pocket and interact favorably with the surrounding protein residues.

3. Structure-based optimization: The ligand binding pocket serves as a target for structure-based optimization. By analyzing the interactions between the ligand and the pocket, medicinal chemists can modify the ligand’s chemical structure to improve its binding affinity, selectivity, and pharmacological properties. This optimization process can involve modifications such as introducing functional groups, adjusting molecular properties, or exploring different scaffolds.

4. Drug selectivity and off-target effects: Understanding the ligand binding pocket’s characteristics is crucial for achieving drug selectivity. By designing ligands that fit precisely into the binding pocket of the target protein, researchers can increase the specificity of the drug and reduce off-target effects. This specificity is essential for minimizing potential side effects and improving therapeutic outcomes.

In summary, the ligand binding pocket is of paramount importance in the structure-based design of small molecule drugs. Understanding the structural and chemical characteristics of the binding pocket, along with employing computational techniques and structural biology methods, enable the identification and optimization of ligands that can bind selectively and with high affinity, ultimately facilitating the development of effective therapeutic interventions.

Structural analysis of small molecule drugs

Molecular structure plays a crucial role in drug development, as the structural features of a molecule determine its ability to interact with biological targets and produce desired therapeutic effects. There are two main types of drugs: small molecules and biologics. Small molecules are typically synthesized chemically and are generally less complex than biologics. They are designed to target specific molecules involved in disease processes and often have a defined structure that allows them to bind to these targets with high specificity.^26,27 By contrast, biologics are typically large, complex molecules, such as proteins, that are produced by living cells and are designed to interact with specific receptors or pathways in the body.²⁷

One important approach to drug discovery is structure-based drug design (SBDD), which uses computational tools to predict the position of small molecules within a 3D representation of the protein structure and estimate the affinity of ligands to target protein with considerable accuracy and efficiency.^28,29 SBDD can also help identify potential off-target effects of a drug candidate, which is important for minimizing unwanted side effects.

The molecular structure of a drug is a critical factor in its ability to interact with biological targets and produce therapeutic effects. Small molecules with specific structural features can be designed to bind to target proteins with high specificity, whereas biologics are typically large, complex molecules that are designed to interact with specific receptors or pathways in the body. For example, G protein-coupled receptors (GPCRs) are a family of cell membrane proteins that are involved in many physiological processes and are important drug targets. Approximately 350 non-olfactory members of the human GPCR family are considered druggable, of which 165 are validated drug targets.³⁰ Small molecules that target GPCRs can have a variety of structural features, such as the presence of a benzene ring or a carboxyl group, which allow them to interact with specific binding sites on the receptor and modulate its activity. Computational tools such as SBDD have been used to design and optimize drug candidates targeting GPCRs, and demonstrating the structural features is important for drug efficacy.^31–33

Functional groups are one of the key structural features that are commonly used in small molecule drug design (Figure 2). In particular, the presence of certain functional groups such as hydroxyl (-OH), carboxyl (-COOH), and amino (-NH₂) groups can enable small molecules to interact with specific enzymes or receptors and confer specific pharmacological properties in the body.^34,35 For example, the presence of hydroxyl groups in molecules such as ethanol can enable them to interact with receptors in the brain, leading to the well-known effects of alcohol consumption.³⁶ Another critical aspect of small molecule drug design is stereochemistry (Figure 2), which refers to the 3D arrangement of atoms in a molecule and how it can affect the molecule’s interaction with biological targets.³⁷ For example, the active form of the anti-inflammatory drug ibuprofen has a specific stereochemistry that allows it to bind to and inhibit the activity of cyclooxygenase enzymes.³⁸ By contrast, the inactive form of ibuprofen lacks this stereochemistry and does not bind to cyclooxygenases.³⁸ Overall, functional groups and stereochemistry are both key features in the design of small molecules. Understanding how different functional groups and stereoisomers can affect a molecule’s interaction with biological targets can help researchers develop more effective and specific drugs for treating a variety of diseases.

FIGURE 2. Key structural features considered in small molecule drug design.

These features include functional groups (e.g., hydroxyl, carboxyl, and amino groups) that give drugs specific pharmacological properties, stereochemistry that can affect a drug’s interaction with biological targets, molecular weight, size and shape that impact pharmacokinetics and pharmacodynamics, and specific chemical bonds and electrostatic properties that influence a drug’s physical and chemical properties. Understanding these features is crucial for designing effective small molecule drugs with desired pharmacological properties.

Molecular weight is also an important consideration in small molecule drug design, as it can affect the pharmacokinetics and pharmacodynamics of a drug. For example, drugs with low molecular weight can be rapidly absorbed and excreted from the body, whereas drugs with high molecular weight may have a longer half-life and require lower dosages.³⁹ Other structural features include the size and shape of the molecule, the presence of specific chemical bonds, and the electrostatic properties of the molecule (Figure 2). The presence of specific chemical bonds in a molecule can affect its structure and properties. For example, covalent bonds are formed when two atoms share a pair of valence shell electrons between them, and atoms of Groups IV through VII bond so as to complete an octet of valence shell electrons.⁴⁰ The electrostatic properties of a molecule, such as its dipole moment and charge distribution, can affect its interactions with other molecules and its reactivity.⁴⁰ The size and shape of a molecule can affect its physical and chemical properties. For example, larger molecules tend to have higher boiling points and melting points than smaller molecules.⁴¹ By understanding the importance of these structural features, researchers can design small molecule drug candidates with optimized pharmacological properties and reduced side effects.

Computational approaches to small molecule drug design

Computational methods play a crucial role in designing new drug candidates and optimizing their properties. The use of computational methods in drug design has increased rapidly in recent years, and numerous tools are available to support this process. Here, we discuss some of the most commonly used computational approaches in small molecule drug design (Figure 3).

FIGURE 3. The major computational approaches used in small molecule drug design.

Molecular docking involves predicting the binding affinity and orientation of small molecules with larger macromolecular targets. Virtual screening is a cost-effective and time-saving method that helps researchers to narrow down the number of compounds for further experimental analysis. Molecular dynamics (MD) simulations provide detailed insights into the physical movements and interactions of atoms and molecules in a system. Quantum mechanics/molecular mechanics calculations combine the accuracy of quantum mechanical calculations with the speed of molecular mechanics simulations to study chemical and biochemical systems. Machine learning algorithms use computational approaches to predict the properties of potential small molecule drug candidates. These computational approaches are widely used in drug discovery and lead optimization to identify potential drug targets and predict molecular ligand-target interactions at the atomic level.

Molecular docking

One of the primary computational methods used in small molecule drug design is molecular docking. Molecular docking involves predicting the binding affinity and orientation of small molecules, known as ligands, with larger macromolecular targets, such as proteins or enzymes.⁴² Docking algorithms use energy minimization and scoring functions to evaluate the stability of the ligand–protein complex and rank potential drug candidates. The process of molecular docking helps researchers to gain insights into the molecular mechanisms of various biological processes such as drug–protein interactions, protein–protein interactions, and enzymatic reactions.^43,44 In molecular docking, ligands are docked into the binding site of the protein, and the resulting complex is scored based on its fitness or complementarity.⁴⁵ The docking process generates multiple possible conformations and orientations of the ligand with the binding site of the target protein.⁴² The goal of molecular docking is to identify the best-fit ligand–protein complex based on the calculated binding affinity and energy of the complex.⁴²

Molecular docking has a wide range of applications in drug discovery, lead optimization, and structure-based drug design.⁴² It allows researchers to identify potential drug targets and predict molecular ligand–target interactions at the atomic level, providing crucial information for drug development.⁴⁴ Moreover, molecular docking can also be used to study food proteins and bioactive peptides, providing insights into their structural and functional properties.⁴³ Together, molecular docking is a powerful computational technique that is widely used in structural biology and drug discovery. It helps researchers gain insights into the molecular mechanisms of various biological processes and identify potential drug targets. The use of molecular docking is expected to continue to grow in the future as more researchers explore its potential applications in various fields.

Over the past few years, remarkable strides have been made in the field of molecular docking, resulting in substantial enhancements in the precision and effectiveness of this technique. For example, (i) Alchemical free energy methods, such as free energy perturbation and thermodynamic integration, have been developed to calculate the binding free energy of a ligand to a protein.⁴⁶ These methods can provide a more accurate prediction of binding affinity than traditional docking methods. (ii) New docking tools, such as HADDOCK and RosettaDock, have been developed to incorporate protein flexibility into docking simulations.^47,48 These tools can predict the conformational changes of a protein upon ligand binding, which can improve the accuracy of docking predictions. (iii) ML has been incorporated into docking tools to improve their predictive accuracy. For example, one study proposed a deep neural network model with an attention mechanism to improve the prediction accuracy of protein-ligand complex binding affinity.⁴⁹ Another study investigated the use of random forest regression as an alternative to traditional linear regression methods and demonstrated improved prediction performance.⁵⁰ These studies highlight the potential of ML techniques, including random forest regression, for predicting binding affinities in protein–ligand interactions. By training on known ligand–protein complexes with experimentally measured binding affinities, ML models can learn patterns and make predictions on new, unseen complexes.

Virtual screening

Another approach that has gained popularity in recent years is virtual screening, which involves screening large compound libraries for potential drug candidates using molecular docking or other techniques. It is a cost-effective and time-saving method that helps researchers narrow down the number of compounds for further experimental analysis. Virtual screening is used to identify small molecules that are likely to bind to a target protein.⁵¹ It involves the use of various software tools such as GOLD, grid-based ligand docking with energetics (GLIDE), and Autodock Vina. Autodock Vina is freely accessible and provides good results for screening different ligands.⁵² In the context of drug discovery, virtual screening can be used to identify chemical structures that have particular properties.⁵³ This process helps researchers identify potential drug candidates that can selectively interact with a target protein while minimizing the side effects. Virtual screening has emerged as a groundbreaking technique that is helping to significantly improve and speed up the process of drug discovery.⁵⁴ Research has shown that virtual screening is effective in scanning the potential affinity of millions of compounds to selected targets simultaneously.

The virtual screening process involves three main steps: preparation of the target protein structure, preparation of the compound library, and the screening of compounds against the target protein. During the preparation of the target protein structure, the protein is optimized by removing any water molecules or co-crystallized ligands. In the preparation of the compound library, a large database of small molecules is created from which potential lead compounds are identified. Finally, during the screening of compounds against the target protein, the compounds are ranked based on their predicted binding affinity to the target protein.⁵⁵ Virtual screening is a powerful tool for identifying potential drug candidates, but it also has limitations. The accuracy of virtual screening results depends on the quality of the protein structure and the compound library. Additionally, the virtual screening process does not consider the pharmacokinetic and toxicological properties of the compounds. Therefore, compounds identified through virtual screening need to be validated experimentally to determine their efficacy and safety.^56,57 In summary, virtual screening is a valuable computational technique that is helping to accelerate drug discovery. It involves the use of software tools to identify potential lead compounds from a large database of molecules. However, virtual screening results need to be validated experimentally, and the limitations of the method need to be considered.

MD simulations

MD simulations are a powerful computational tool used to study the physical movements and interactions of atoms and molecules in a system.⁵⁸ MD simulation integrates Newton’s equations of motion over time to obtain the motion of the atoms/molecules in a system, which provides quantitative and qualitative information about the macroscopic behavior of the system at the atomic level.⁵⁸ In other words, MD simulations provide a ‘movie’ of the dynamic ‘evolution’ of the system under investigation.

MD simulations consist of the numerical, step-by-step, solution of the classical equations of motion, which for a simple atomic system may be written as m_i * a_i = f_i, where m_i is the mass of particle i, a_i is its acceleration, and f_i is the force acting on particle i.⁵⁹ The forces acting on each particle are determined by the interatomic interactions, which are typically represented by a mathematical function that describes the potential energy of the system as a function of the atomic positions. The interatomic potential function can be obtained using quantum mechanics, empirical potentials, or a combination of both.⁵⁹ Figure 4 illustrates the basic steps and principle of working of MD simulations.

FIGURE 4. The basic steps and principle of working of molecular dynamics (MD) simulations used in new drug development.

(a) Schematic representation of the steps involved in MD simulation for drug research. This figure illustrates the sequential steps in conducting MD simulations for drug research. The setup and preparation phase involves obtaining and converting the three-dimensional structures of protein and drug molecules into appropriate file formats. Essential simulation parameters, such as temperature, pressure, and time step, are defined. Additionally, an appropriate force field is chosen based on the system and research objectives. It is important to select simulation software that is compatible with the chosen force field. (b) During the simulation process, energy minimization is performed to alleviate steric clashes and minimize potential energy in the system. The equilibration phase allows the system to gradually relax by restraining specific atoms and allowing solvent molecules to adjust around the protein–drug complex. Subsequently, a time-dependent molecular dynamics run is carried out, where the equations of motion are numerically integrated to generate atom trajectories. (c) The analysis of the topology file includes various metrics. Root-mean-square deviation (RMSD) is used to measure the average deviation between different structures at different time points and a reference structure. Root-mean-square fluctuation (RMSF) determines the atomic fluctuations of the protein and drug, providing insights into their flexibility and stability. The radius of gyration (Rg) quantifies the compactness or spatial extent of the protein–drug complex. Hydrogen bonds are assessed to understand their formation and dynamics, indicating potential interactions. Finally, the solvent accessible surface area (SASA) is calculated to analyze the interactions of the protein and drug with the surrounding solvent molecules.

MD simulations can be used to study a wide range of systems, including biological macromolecules, materials science, and chemical physics.⁶⁰ When applied to biological macromolecules, MD simulations can provide insights into the dynamic behavior of proteins and nucleic acids, including fluctuations in the relative positions of the atoms in a molecule.⁶¹ MD simulations are important in the theoretical study of food proteins and bioactive peptides, and can be used in conjunction with molecular docking to predict the binding of small molecules to proteins.⁴³ MD simulations can be performed using different levels of detail, from atomistic simulations that simulate individual atoms to coarse-grained approaches that lump a number of atoms into pseudo-particles.^62,63 The choice of simulation method depends on the level of detail required for the system under investigation.

Taken together, MD simulations are a powerful computational tool that can provide detailed insights into the physical movements and interactions of atoms and molecules in a system. The simulations integrate Newton’s equations of motion over time to obtain the motion of the atoms/molecules in a system and can provide quantitative and qualitative information about macroscopic behavior of the system at the atomic level. MD simulations can be used to study a wide range of systems, from biological macromolecules to materials science, and can be performed at different levels of detail depending on the system under investigation.

Recently, the following noteworthy advancements have emerged in the field of MD simulations. (i) Deep learning algorithms increasingly applied to MD simulations. For instance, Atomic Convolutional Networks have been used to predict the potential energy of a molecule directly from its atomic coordinates, which can significantly improve the accuracy of MD simulations.⁶⁴ (ii) Enhanced sampling techniques, such as metadynamics, have been developed to overcome the limitations of traditional MD simulations.^65,66 These techniques can explore a larger conformational space and provide a more accurate representation of the energy landscape of a system. (iii) Coarse-grained models, which simplify the representation of a system by grouping atoms together, have been developed to reduce the computational cost of MD simulations.^67,68 These models can simulate larger systems and longer timescales than traditional all-atom models. (iv) GPU-accelerated tools, such as GROMACS and AMBER, have been developed to speed up MD simulations.^69,70 These tools can perform simulations faster than CPU-based tools, enabling the study of larger and more complex systems. Together, these advancements have significantly improved the accuracy and efficiency of MD simulations. However, challenges remain, such as the accurate prediction of protein flexibility and the computational cost of large-scale simulations. The field is rapidly evolving, and we can expect further improvements in the coming years.

Quantum mechanics/molecular mechanics calculations

Quantum mechanics/molecular mechanics (QM/MM) calculations are a powerful computational approach used to study chemical and biochemical systems at the atomic and molecular level. The QM/MM approach combines the accuracy of quantum mechanical calculations with the speed of molecular mechanics simulations.⁷¹ This method allows for the calculation of thermodynamic properties and the characterization of reaction dynamics, making it a valuable tool for studying chemical and biochemical systems in solution or enzymes.⁷² QM/MM calculations can provide accurate calculations of reaction energies and reaction pathways.

In traditional molecular mechanics and quantum mechanics computations, inter- and intramolecular interactions are evaluated for a given frozen configuration of the system, often without explicit solvation.⁷³ However, the QM/MM approach can take solvation effects into account by treating the QM region as a solute and the MM region as a solvent.⁷⁴ The QM/MM approach was first introduced in 1976 by Warshel and Levitt.⁷⁵ Since then, it has been widely used to study a variety of chemical and biochemical systems including enzymes, reaction mechanisms, and protein–ligand binding interactions. One of the advantages of the QM/MM approach is its ability to simulate larger systems than can be treated with quantum mechanical ab initio methods alone.⁷⁴ The size of many systems of interest in chemistry and biochemistry prevents efficient and accurate treatment by quantum mechanical methods alone. Overall, the QM/MM approach is a valuable tool for studying the behavior of matter and light at the atomic and subatomic scale, making it a useful approach for understanding the properties of molecules and atoms and their constituents.⁷⁶

ML algorithms

ML algorithms have become increasingly popular in predicting the properties of small molecules. These algorithms use computational approaches to predict the properties of potential small molecule drug candidates such as their solubility, toxicity, and binding affinity.⁷⁷ Two main groups of ML methods are discussed in drug discovery: traditional ML methods (e.g., tree-based methods, latent variable methods) and deep learning methods.⁷⁸ The characterization of molecular properties is a critical problem in drug discovery. Experimental methods have been widely used across the entire drug discovery process, including high-throughput in vitro screening and low-throughput in vivo testing. However, on average, one United States Food and Drug Administration (FDA) drug is approved for five compounds entering clinical trials, which shows the need for more efficient drug discovery methods.⁷⁹

ML approaches vary in complexity and range from simple sum-over-atoms methods to more sophisticated approaches capable of describing collective interactions between many atoms or bonds.⁸⁰ For example, ML models have been used to predict the binding of small molecules to RNA targets, and Lasso regression models were used to compare the performance of various ML algorithms to predict the binding scores of phenylthiazole-containing molecules.⁸¹

Solubility prediction, which can reduce waste and improve the crystallization process efficiency, has attracted increasing attention. However, there are still many urgent challenges, and several methods are being developed to address them.⁸² ML models, such as MegaTox, can be used to predict early-stage clinical compounds and recent FDA-approved drugs to identify potential drug-induced liver injury.⁸³

Overall, ML algorithms have shown great promise in predicting the properties of small molecule drug candidates. These algorithms have the potential to revolutionize drug discovery by reducing the need for expensive experimental methods, making drug development faster and more efficient.

Computational approaches have made remarkable strides in advancing the field of small molecule drug design; however, these approaches are not without their inherent limitations. Within this context, several key challenges deserve careful consideration, including accounting for protein flexibility, improving scoring functions, addressing solvent effects, extending simulation time scales, managing computational costs, acquiring adequate experimental data, predicting induced fit, and accurately anticipating off-target effects.

1. Protein flexibility: Traditional docking algorithms often treat the protein as a rigid body, which is a significant oversimplification. Proteins are highly dynamic and can adopt a range of conformations. Accounting for this flexibility is a major challenge for docking algorithms.

2. Scoring functions: The scoring functions used in docking algorithms to predict the binding affinity of a ligand to a protein are often inaccurate. They are typically based on simplified models of molecular interactions and do not fully capture the complexity of these interactions.

3. Solvent effects: Many docking algorithms do not adequately account for the effects of the solvent. The solvent can play a crucial role interactions, and neglecting these effects can lead to inaccurate predictions.

4. Time scale: MD simulations can model the dynamic behavior of molecules, but they are limited by the time scale they can simulate. Most simulations are on the nanosecond to microsecond scale, while many biological processes occur on the millisecond to second scale.

5. Computational cost: Both docking and MD simulations are computationally intensive. This limits the size of the systems that can be studied and the length of the simulations that can be performed.

6. Lack of experimental data: There is often a lack of experimental data to validate the predictions made by docking and MD simulations. This makes it difficult to assess the accuracy of these methods.

7. Difficulty in predicting induced fit: Many docking algorithms struggle to predict the conformational changes that occur when a ligand binds to a protein, a phenomenon known as induced fit.

8. Limitations in predicting off-target effects: While computational methods are improving in predicting the interaction between a drug and its intended target, predicting potential off-target effects remain a significant challenge. Despite these challenges, the computational approaches used in small molecule drug design are continuously evolving. Emerging computational tools and techniques such as deep learning and artificial intelligence are likely to have a significant impact on drug design in the future.⁸⁴ These tools can help researchers generate new hypotheses, analyze large data sets, and design new drug candidates with enhanced efficacy and safety profiles.

In conclusion, computational methods play a vital role in small molecule drug design. Molecular docking and virtual screening are commonly used methods for predicting ligand–protein interactions and screening large compound libraries. Other computational approaches such as MD simulations, quantum mechanics/molecular mechanics calculations, and ML algorithms have also been applied to drug design. Emerging computational tools and techniques are likely to further advance the field of small molecule drug design in the future.

Case studies

In recent years, there have been several notable case studies that highlight the importance of molecular structure in small molecule drug design. These case studies demonstrate the successful application of structure-based drug design and provide insights into the discovery process, mechanism of action, and pharmacological properties of the drugs. One such example is venetoclax, a small molecule drug used for the treatment of chronic lymphocytic leukemia (CLL) and small lymphocytic lymphoma (SLL).⁸⁵ Venetoclax was designed based on the identification of the anti-apoptotic protein B-cell lymphoma 2 (BCL-2) as a therapeutic target.⁸⁶ The drug’s chemical structure consists of a bicyclic scaffold with a 4-aminoindole core and two benzenesulfonamide moieties.^85,87 Venetoclax binds selectively to BCL-2, leading to apoptosis of cancer cells by blocking the interaction between BCL-2 and pro-apoptotic proteins.^85,88 Venetoclax has been shown to be effective in the treatment of CLL and SLL, with a favorable safety profile.^86,89 Its pharmacological properties include a long half-life of approximately 26 h and linear pharmacokinetics with dose-proportional exposure.⁹⁰ The discovery of venetoclax was made possible by the application of structure-based drug design, which involves the use of computational and experimental techniques to design small molecules that can selectively bind to a target protein.^87,91,92 The structure of BCL-2 was determined by X-ray crystallography, which allowed for the identification of the binding site and the rational design of Venetoclax.^93,94 Molecular docking and other computational methods were used to optimize the structure of venetoclax and predict its binding affinity.⁹⁵

Another example is the development of the drug ivacaftor, which is used to treat cystic fibrosis. Ivacaftor was designed based on its ability to selectively bind to and enhance the activity of the cystic fibrosis transmembrane conductance regulator (CFTR) protein, which is defective in patients with cystic fibrosis. The drug was designed to interact with specific amino acid residues in the CFTR protein, stabilizing the protein and increasing its activity.⁹⁶ Once the drug target (CFTR) is chosen, the 3D structure of the protein is determined using techniques such as X-ray crystallography or NMR spectroscopy. The protein structure is then used to identify potential binding sites for small molecules.⁹⁷ In the case of ivacaftor, the binding site was located in the intracellular domain of the CFTR protein. The next step is to design small molecules that can interact with the target protein and modify its function. This is done using computer-aided drug design (CADD) tools such as molecular docking and MD simulations.⁹⁸ These tools allow researchers to predict the position of small molecules within a 3D representation of the protein structure and estimate the affinity of ligands to target protein with considerable accuracy and efficiency.²⁹ After the small molecule is designed, it undergoes a series of tests to determine its efficacy and safety. This includes in vitro assays, in vivo studies, and toxicity tests. If the drug is found to be effective and safe, it proceeds to clinical trials.^99,100 In brief, ivacaftor was developed using the structure-based drug design methodology. The process involves the choice of a drug target, determination of the protein structure, identification of potential binding sites, design of small molecules using CADD tools, and testing of the drug for efficacy and safety.

A recent case study involves the development of a novel metabolic reprogramming strategy (MRS) for the treatment of diabetes-associated breast cancer.¹⁰¹ As part of this strategy, we aimed to identify potential monocarboxylate transporter 4 (MCT4) inhibitors for combination therapy. The 3D model of human MCT4 structure was generated by using the I-TASSER On-line Server. Then, we used grid-based ligand docking with energetics (GLIDE) software to virtually screen seven commercial compound libraries and identify potential MCT4 inhibitors. The top-ranked compounds were visually inspected, and the most promising candidate CB-2 was pursued for further testing. Experimental testing confirmed the binding of CB-2 to MCT4, and further studies demonstrated its cytotoxic activity against cancer cells and its antitumor effects on animal models. The methodology used in this study highlights the importance of structure-based drug design in the development of novel small molecule drug candidates. At present, CB-2 has obtained a global patent and entered the preclinical study stage.

In addition to the above few examples, there are some classic case studies, such as the development of the drug olaparib for the treatment of breast and ovarian cancers, and the development of the drug sofosbuvir for the treatment of hepatitis C. Olaparib was designed based on its ability to selectively inhibit poly(ADP-ribose) polymerase (PARP), an enzyme that repairs damaged DNA in cancer cells. The drug was designed to mimic the structure of the substrate of PARP, allowing it to selectively bind to and inhibit the enzyme in cancer cells.^102,103 Sofosbuvir was designed based on its ability to selectively inhibit the RNA polymerase of the hepatitis C virus. The drug was designed to mimic the structure of the natural substrate of the polymerase, allowing it to selectively bind to and inhibit the enzyme in the virus.¹⁰⁴ Overall, these case studies emphasize the significance of molecular structure in drug design and its contribution to the development of effective therapeutic options.

Concluding remarks and future directions

Small molecule drug design is a rapidly growing field that has the potential to revolutionize the development of new pharmaceuticals. As advancements in structural biology and computational methods continue to expand our understanding of the relationship between molecular structure and drug efficacy, researchers are finding new and innovative ways to design small molecules that are more potent, selective, and safe than ever before. In this review, we explored some of the recent advancements in small molecule drug design from a structural perspective and discussed the emerging trends and future directions in this field.

One of the key areas of research in this field is the development of new computational tools and methods for predicting and optimizing the structural properties of small molecules. These tools are becoming increasingly important as the size and complexity of drug targets continue to grow, making it more difficult to design drugs using traditional methods. Some of the emerging computational methods that are being used in small molecule drug design include ML algorithms, MD simulations, and quantum mechanical calculations.¹⁰⁵ Besides, other computational tools and methods such as statistical models have been developed for predicting toxicity and side effects of small molecules. These tools enable the identification and optimization of lead compounds, and the prediction of their pharmacological properties.^87,106 Overall, the development of new computational tools and methods is crucial for the efficient and cost-effective discovery and development of small molecule drugs. These tools enable the prediction and optimization of the structural properties of small molecules, which is critical for their safety and efficacy. With the increasing use of ML and other computational methods in drug discovery, the future of small molecule drug design looks promising.⁸⁴

Another important trend in small molecule drug design is the exploration of new therapeutic applications. While traditional drug design has focused primarily on treating common diseases such as cancer and cardiovascular disease, researchers are now beginning to investigate the potential of small molecule drugs in treating rare and neglected diseases. This has led to the development of new drug discovery programs focused on diseases such as rare genetic disorders, neglected tropical diseases, and emerging infectious diseases.^107–109

Finally, the future of small molecule drug design is likely to be shaped by new technologies and innovations in the field of structural biology. For example, the recent development of cryo-electron microscopy has revolutionized the way that researchers study the structures of biological macromolecules, and has already led to the discovery of new drug targets and the development of new drugs.^110,111 Other emerging technologies that are likely to have an impact on small molecule drug design include protein engineering, chemical genomics, and high-throughput screening.^112,113

In conclusion, small molecule drug design is a rapidly evolving field that is driven by advancements in structural biology and computational methods. As researchers continue to explore the relationship between molecular structure and drug efficacy, new and innovative small molecule drug candidates are being developed that have the potential to revolutionize the treatment of a wide range of diseases. Looking to the future, the development of new computational tools and methods, the exploration of new therapeutic applications, and the continued evolution of structural biology are likely to be key drivers of innovation in this field.

Teaser:

Exploring the cutting-edge advancements in small molecule drug design, delving into the role of protein structures, computational tools, and optimization techniques. This review illuminates how integrating structural data can revolutionize drug discovery, enhancing efficacy, specificity, and safety.

Research Highlights:

1. Importance of accurate protein structure prediction for small molecule drug design.

2. Key structural features and properties used in small molecule drug design.

3. Computational tools and methods for predicting and optimizing small molecule structures.

4. Examples of successful small molecule drug design based on molecular structure.

Abbreviations

BCL-2

B-cell lymphoma 2

CADD

computer-aided drug design

CFTR

cystic fibrosis transmembrane conductance regulator

CLL

chronic lymphocytic leukemia

cryo-EM

cryo-electron microscopy

GLIDE

grid-based ligand docking with energetics

GPCRs

G protein-coupled receptors

MCT4

monocarboxylate transporter 4

MD

molecular dynamics

ML

machine learning

MRS

metabolic reprogramming strategy

PARP

poly(ADP-ribose) polymerase

QM/MM

quantum mechanics/molecular mechanics

SBDD

structure-based drug design

SLL

small lymphocytic lymphoma

Data availability

No data were used for the research described in the article.