Advancements in Molecular Docking: A Comprehensive Review of Methods and Their Applications in Drug Discovery

Molecular docking is an important computer-based method used in modern drug discovery. It helps scientists find and improve new medicines by studying how drug molecules fit and interact with their target proteins. This technique plays a key role in speeding up and simplifying the process of developing new drugs.

Overview of Molecular Docking:

Molecular docking is a computer-based method used to predict how small molecules (ligands) attach to larger molecules like proteins. It helps understand how they fit together and form stable, energy-efficient complexes. The process involves two key stages: a conformational search that explores possible ligand positions and orientations within the protein’s binding site, and a scoring phase? that evaluates these poses based on binding affinity using factors such as geometric complementarity, electrostatic and van der Waals interactions, hydrogen bonding, hydrophobic effects, and desolvation energy¹.

Docking is based on shape and chemical complementarity, illustrated by the “lock-and-key” and “induced-fit” models. In the induced-fit model, both the ligand and protein adjust their shapes to form a stable, low-energy complex. This provides detailed insight into molecular interactions, helping researchers understand drug–target binding, identify key residues, and explain structure–activity relationships².

Historical Development and Evolution:

Molecular docking has evolved greatly since the early 1980s. The first program, DOCK (1982), introduced by Kuntz and colleagues, used simple rigid-body models based on geometric matching. In the 1990s and early 2000s, semi-flexible docking methods were developed, allowing ligand flexibility while keeping proteins rigid, with tools like FlexX, GOLD, and AutoDock improving accuracy. Later, fully flexible docking incorporated movements of both ligands and proteins using ensemble docking and molecular dynamics for more realistic modeling. Recently, AI and machine learning have further advanced docking by improving conformational sampling and scoring, leading to faster and more accurate prediction of binding affinities and large-scale virtual screening³.

Importance in Modern Drug Discovery:

Molecular docking plays a crucial role in today’s drug discovery by supporting structure-based design, virtual screening, and lead optimization. In structure-based design, it helps identify and design compounds that fit precisely into target binding sites, improving hit rates and reducing unwanted interactions. Through virtual screening, docking enables the rapid evaluation of millions of compounds to find promising candidates more efficiently than traditional methods. In lead optimization, it assists in refining hits to enhance binding strength, selectivity, and pharmacological properties.Molecular docking provides faster ,cost-effective, and ethical alternative to experimental approaches, greatly accelerating the discovery of potent and specific drug molecules⁴.

Scope and Organization of the Review:

This review covers both the theoretical and practical aspects of molecular docking, including its basic principles, binding energetics, and computational algorithms. It discusses key docking methods, search strategies, scoring functions, and essential steps such as ligand preparation, protein flexibility, and validation. Advanced applications like fragment-based design, covalent and peptide docking, and protein–protein interaction modeling are also highlighted. Emerging trends such as AI-driven docking, ultra-large library screening, and allosteric site targeting are explored, emphasizing the expanding role of molecular docking in next-generation drug discovery⁵.

Fundamental Principles of Molecular Docking:

Molecular docking rests upon well-established physicochemical and thermodynamic principles that govern molecular recognition and binding between biological macromolecules and small molecule ligands. Understanding these fundamental concepts is essential for appreciating both the theoretical foundation and practical application of docking methodologies in drug discovery.

Lock-and-Key vs. Induced-Fit Models:

The understanding of molecular recognition has evolved from the classical lock-and-key model to the more dynamic induced-fit model.Modern research further suggests a balance between conformational selection and induced fit, showing that proteins exist in multiple conformations even before binding. In molecular docking, this theoretical shift has influenced algorithmic development—from early rigid-body (lock-and-key) methods to flexible and dynamic docking approaches that account for protein movement and ligand-induced conformational changes. This evolution has enhanced docking accuracy and better reflects real biological interactions⁶.

Fig. 1: Lock-and-Key vs Induced-Fit Mechanisms of Enzyme Activity

Molecular recognition represents the fundamental process by which a ligand specifically identifies and binds its target protein through complementary interactions. This specificity arises from the integration of multiple forms of complementarity geometric, electrostatic, hydrophobic, and hydrogen-bonding that collectively determine binding selectivity and affinity⁷.

Shape complementarity is a key factor in molecular recognition, describing how well a ligand fits into a protein’s binding pocket. Optimal fit minimizes steric clashes and maximizes surface contact, enhancing binding affinity. Studies show that high shape complementarity strongly predicts binding strength and occurs consistently across various protein-ligand systems, establishing geometric fit as a universal principle of molecular interaction⁸.

Physicochemical complementarity encompasses electrostatic, van der Waals, hydrophobic, and desolvation interactions that collectively drive and stabilize ligand-receptor binding.Van der Waals interactions, though weak individually, collectively stabilize ligand-protein binding by ensuring optimal atomic contact distances. These forces prevent steric clashes and excessive gaps, complementing shape fit and reinforcing overall molecular recognition⁹.

Hydrophobic interactions drive ligand-protein binding by clustering nonpolar regions and displacing water from binding sites. This effect provides significant binding energy and stability, with hydrophobic complementarity strongly correlating with binding affinity and occurrence across diverse protein systems¹⁰.

Hydrogen bonding between donor and acceptor groups on ligands and proteins provides additional specificity and binding energy that enhancing selectivity beyond geometric and hydrophobic effects¹¹. Hydrogen bonding enhances ligand-protein specificity and stability by forming directional interactions that anchor the complex.Desolvation plays a key role in molecular recognition, as water displacement during binding affects overall energy. Effective ligands balance favorable hydrophobic burial with minimal desolvation penalties, promoting stable and efficient binding¹².

Binding Free Energy:

The thermodynamic stability and spontaneity of ligand-receptor complex formation are quantitatively described by binding free energy, a concept central to both understanding and predicting protein-ligand interactions in molecular docking. Gibbs free energy change (ΔG) represents the fundamental thermodynamic parameter governing binding thermodynamics, expressing the energy available to drive ligand-protein complex formation¹³.

Ligand binding is governed by enthalpic and entropic contributions. Entropy involves conformational and solvation changes; although binding restricts molecular motion driving stability. Docking scoring functions estimate binding free energy using simplified models that combine van der Waals, electrostatic, hydrogen bonding, and desolvation terms¹⁴.

Molecular Docking Methodologies:

Molecular docking approaches vary widely to balance computational efficiency and biological accuracy. They are generally classified based on treatment of molecular flexibility, sampling algorithms for conformational exploration, and the strategies used for pose prediction and scoring.

Classification Based on Flexibility:

Molecular docking methods differ in how they handle flexibility, ranging from fast rigid models to more accurate fully flexible approaches.

Rigid Docking:

Rigid docking is a computational technique in which both the ligand and receptor are considered fixed three-dimensional structures, allowing only translational and rotational movements within six degrees of freedom¹⁵.

The main advantage of rigid docking is its high computational efficiency. Due to the limited conformational search space, it enables rapid screening of large compound libraries and is therefore effective for primary filtering in high-throughput virtual screening workflows¹⁶.

A key limitation of rigid docking is its inability to represent conformational flexibility in either ligand or protein. It fails to capture induced-fit effects or side-chain rearrangements that occur during binding, resulting in reduced predictive accuracy. Consequently, important interactions such as hydrogen bonding or accommodation of bulky side chains are often missed¹⁷.

Semi-Flexible Docking:

Semi-flexible docking, also known as flexible ligand/rigid receptor docking, treats the ligand as flexible while keeping the receptor rigid. This approach provides a balance between computational efficiency and predictive accuracy, making it the most widely applied strategy in modern structure-based drug discovery¹⁸.

Semi-flexible docking offers higher accuracy than rigid docking by accounting for ligand flexibility. It is implemented in major docking tools such as AutoDock, FlexX, Glide, and Surflex, employing algorithms like Monte Carlo simulated annealing and genetic optimization for conformational sampling¹⁹. Despite these improvements, semi-flexible docking assumes a rigid protein structure, reducing accuracy when significant receptor conformational changes occur. Cross-docking studies reveal accuracy loss when ligands are docked into alternative protein conformations, particularly in highly flexible targets such as kinases and proteases²⁰.

Fig. 2: Types of Molecular Docking Based on Structural Flexibility

Flexible docking, or fully flexible docking, allows both ligand and receptor to undergo conformational changes during docking, providing a biologically realistic model of molecular recognition ²¹.

The principal application domains for flexible docking include enzyme inhibition studies where active-site architecture changes substantially upon inhibitor binding, and protein-protein interaction modeling where substantial conformational rearrangement often occurs upon complex formation. Flexible docking proves particularly valuable for kinase inhibitor design, where hinge-bending between the two kinase lobes and loop rearrangements frequently accompany inhibitor binding²².

The limitations of flexible docking include high computational cost and challenges in accurate conformational sampling, as allowing flexibility in protein side chains greatly increases the number of degrees of freedom beyond the ligand’s typical range²³.

Induced-Fit Docking:

Induced-fit docking provides a balanced approach between rigid and fully flexible docking by modelling protein conformational changes during ligand binding through a structured, two-step mechanism that separates ligand sampling from receptor adjustment²⁴ .

In this process, the ligand is first docked into a rigid receptor, followed by receptor side-chain and backbone relaxation to accommodate the ligand, and subsequent redocking and energy minimization. The advanced IFD-MD variant further improves accuracy, reaching an 85% success rate in 258 cross-docking cases. Other implementations, such as Rosetta Ligand, Adaptive BP-Dock, and CHARMM-GUI IFD, employ various strategies like rotamer sampling and solvent-based refinement, achieving up to 80% success in cross-docking predictions²⁵.

Docking Approaches and Algorithms:

Molecular docking aims to identify optimal ligand–receptor binding poses within high-dimensional conformational space using two broad strategies: shape complementarity and simulation-based approaches. Shape complementarity methods rely on geometric matching of ligand and receptor surfaces, using molecular surface definitions and Fourier shape descriptors to identify spatial complementarity without explicit energy calculations²⁶.

In contrast, simulation-based approaches evaluate ligand–protein interaction energies through force fields such as AMBER, CHARMM, and OPLS, explicitly computing van der Waals, electrostatic, and hydrogen-bonding interactions²⁷.

Search Algorithms:

Efficient search algorithms address the challenge of exploring ligand conformational space. Genetic algorithms, as in AutoDock and GOLD, evolve ligand poses through crossover, mutation, and selection, while hybrid Lamarckian implementations integrate local optimization for improved convergence ²⁸.

Specialized Approaches:

Beyond conventional rigid-body and semi-flexible docking paradigms, several specialized approaches address particular challenges in drug discovery. Fragment-based incremental construction in FlexX reduces computational burden by docking rigid fragments sequentially. Gradient-based optimizers, exemplified by AutoDock Vina, refine poses along energy gradients for rapid and accurate docking. Inverse docking screens a ligand against multiple proteins to identify potential off-target interactions and repurposing opportunities using normalized Z-score rankings ²⁹.

Scoring Functions:

Scoring functions are mathematical models that evaluate protein-ligand interactions by converting 3D docking poses into numerical values reflecting binding affinity. They are crucial for distinguishing favorable binding modes and directly determine the accuracy and reliability of molecular docking predictions.

Types of Scoring Functions:

Molecular docking scoring functions can be classified into three major categories reflecting their underlying theoretical frameworks and derivation methodologies.

1.Force Field-Based Scoring Functions:

Force field-based scoring functions estimate binding free energy using classical molecular mechanics formulations that sum van der Waals, electrostatic, solvation, and conformational entropy terms³⁰.

ΔGbind=ΔEvdW+ΔEelec+ΔGsolv+ΔSconf

The van der Waals interactions, modeled via the Lennard-Jones potential, describe nonbonded contacts that favor optimal atomic packing while penalizing steric clashes or excessive separations. Electrostatic contributions are computed using Coulomb’s law with a distance-dependent dielectric function to capture solvent screening effects efficiently, while more sophisticated implementations use Poisson–Boltzmann or Generalized Born models for improved electrostatic accuracy. Solvation energy is generally estimated from solvent-accessible surface area (SASA), capturing hydrophobic interactions and desolvation penalties associated with binding. Prominent examples include DOCK, AutoDock, GOLD/GoldScore, and DockThor, which integrate these physical principles in varying degrees. AutoDock 4.2, for instance, employs a semi-empirical free energy model combining six pairwise potential terms with conformational entropy correction, offering enhanced modeling for halogen and metal ion interactions³¹.

2.Empirical Scoring Functions:

Empirical scoring functions estimate binding free energy by summing weighted terms that represent various interaction contributions, such as van der Waals forces, hydrogen bonding, hydrophobic effects, desolvation, and entropy losses³².

 These functions employ the general form:

ΔGbind=a⋅ΔEvdW+b⋅ΔEhbond+c⋅ΔEhydrophobic+d⋅ΔEentropy+...

The coefficients of these terms are determined by regression fitting to experimental binding affinities, allowing the model to directly capture empirical trends rather than relying solely on physical laws. This approach originated with Böhm’s LUDI scoring function, which established the principle of parameterizing scoring terms using experimental data to improve correlation between predicted and observed affinities. Representative examples include ChemScore, GlideScore, PLP, PLANTS/CHEMPLP, X-Score, and Surflex, many of which introduce additional terms for metal coordination and torsional entropy, optimized through techniques such as ant colony optimization³³. Limitations include the risk of overfitting, dependence on high-quality experimental data, poor transferability across different target classes, and limited treatment of solvation and entropy effects³⁴.

3.Knowledge-Based Scoring Functions:

Knowledge-based scoring functions estimate binding affinity by deriving interaction potentials from statistical analysis of experimentally determined protein–ligand complexes deposited in structural databases such as the Protein Data Bank. These functions calculate mean force potentials based on

W(r)=-RTln?(Nobs(r)Nexp(r))

Knowledge-based methods offer key advantages such as independence from parameter fitting, inherent incorporation of favorable and unfavorable interactions through statistical trends, and broad applicability across diverse targets without retraining. Limitations of knowledge-based scoring functions include dependence on database size and quality, simplified treatment of solvation and entropy effects, and potential biases from uneven structural data representation³⁶.

4.Consensus Scoring:

Consensus scoring combines results from multiple scoring functions to improve ligand ranking accuracy and reduce both false positives and false negatives in virtual screening. It operates on the principle that integrating diverse scoring models provides more reliable predictions than any single function. Two main strategies are used: score-based combinations, which average or weight raw docking scores, and rank-based combinations, which merge compound ranks to overcome the incommensurability of scoring scales³⁷.

 Among advanced implementations, the exponential consensus ranking (ECR) method has shown superior performance by weighting individual scoring functions using exponential distributions. Advantages of consensus scoring include reduced false positive rates through requiring agreement among independent methods, improved discriminatory ability compared to single functions in many benchmarking studies, and parameter-independence in rank-based approaches. Consensus approaches prove particularly effective in virtual screening campaigns where elimination of false positives is critical.?Limitations of consensus scoring include dependence on the quality and diversity of component scoring functions, increased computational cost, and the need for empirical optimization of weighting schemes for specific targets³⁸.

Goals of Scoring Functions:

Scoring functions in structure-based drug discovery have three main goals: predicting the correct binding pose, identifying active compounds in virtual screening, and estimating binding affinity. Pose prediction focuses on reproducing the experimentally observed binding mode. Virtual screening aims to rank true actives within the top fraction of large compound libraries. Binding affinity prediction is the most challenging goal, seeking accurate free energy estimates for potency ranking. Because these goals sometimes conflict, specialized scoring functions are often developed for each task rather than relying on a single universal model³⁹.

Molecular Docking Software and Tools:

Molecular docking software implementing diverse algorithms and scoring functions falls into three categories:

Molecular Docking Workflow and Protocol:

The successful application of molecular docking requires systematic execution of a multi-stage workflow converting raw biomolecular structures into reliable binding predictions. This workflow encompasses protein preparation, ligand preparation, binding site identification, docking execution, and result analysis, each stage critically influencing downstream accuracy and reliability⁴⁰.

General Docking Procedure:

1.Target Protein Preparation:

Target protein preparation involves obtaining structures from the Protein Data Bank or predictive models like AlphaFold, selecting ligand-bound conformations, removing nonessential molecules, adding hydrogens, and performing energy minimization to refine the structure. Nonessential molecules like water, ions, and bound ligands are removed, keeping only those important for binding. Hydrogen atoms are added, and missing residues or bonds are fixed using tools like Schrödinger’s Protein Preparation Workflow or MolProbity⁴¹.

2.Ligand Preparation:

Ligands for molecular docking are sourced from databases like ZINC and PubChem or designed for virtual screening libraries containing thousands to millions of compounds. SMILES or 2D structures are converted into 3D forms and assigned accurate protonation states at physiological pH. Energy minimization and conformer generation optimize ligand geometries for accurate docking. Identifying rotatable bonds helps determine ligand flexibility, which influences docking search complexity and accuracy⁴².

3.Binding Site Identification:

Binding site determination employs static, dynamic, or hybrid approaches depending on structural data availability. When reference ligands are present, their binding locations define target sites; alternatively, computational cavity detection algorithms such as PASS, SurfNet, or POCKET identify potential sites through geometric or energy-based mapping ⁴³.

4.Docking Execution:

Docking execution involves setting up grid parameters to precompute interaction maps, which accelerates docking by replacing real-time energy calculations with grid interpolation. Docking simulations employ search algorithms such as the Lamarckian genetic algorithm in AutoDock 4 and stochastic global search in AutoDock Vina. Parameters like population size, number of runs, and exhaustiveness influence the search depth and docking accuracy⁴⁴.

5.Result Analysis and Validation:

Result analysis uses visualization tools like PyMOL or ChimeraX to evaluate hydrogen bonding, hydrophobic, and electrostatic interactions. Predicted binding energies are used to estimate ligand affinity and rank docking poses for interpretation ⁴⁵.

Fig. 3: Schematic Overview of the Molecular Docking Workflow