Molecular docking and dynamics in protein serine/threonine kinase drug discovery: advances, challenges, and future perspectives

TYPE Review PUBLISHED 27 October 2025 DOI 10.3389/fphar.2025.1696204 OPEN ACCESS EDITED BY Sajjad Gharaghani, University of Tehran, Iran REVIEWED BY

Gianluigi Lauro, University of Salerno, Italy Sirish Kaushik Lakkaraju, Bristol Myers Squibb, United States *CORRESPONDENCE

Anas Shamsi, anas.shamsi18@gmail.com Md. Imtaiyaz Hassan, mihassan@jmi.ac.in RECEIVED 31 August 2025 ACCEPTED 15 October 2025 PUBLISHED 27 October 2025

Molecular docking and dynamics in protein serine/threonine kinase drug discovery: advances, challenges, and future perspectives Gulam Mustafa Hasan 1, Taj Mohammad 2, Sobia Zaidi 3, Anas Shamsi 4* and Md. Imtaiyaz Hassan 2* 1 Department of Basic Medical Science, College of Medicine, Prince Sattam Bin Abdulaziz University, AlKharj, Saudi Arabia, 2Centre for Interdisciplinary Research in Basic Sciences, Jamia Millia Islamia, New Delhi, India, 3Department of Biomedical Sciences, Heritage College of Osteopathic Medicine, Ohio University, Athens, OH, United States, 4Centre of Medical and Bio-Allied Health Sciences Research, Ajman University, Ajman, United Arab Emirates

Hasan GM, Mohammad T, Zaidi S, Shamsi A and Hassan MI (2025) Molecular docking and dynamics in protein serine/threonine kinase drug discovery: advances, challenges, and future perspectives. Front. Pharmacol. 16:1696204. doi: 10.3389/fphar.2025.1696204 COPYRIGHT

© 2025 Hasan, Mohammad, Zaidi, Shamsi and Hassan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Protein serine/threonine kinases (STKs) regulate critical signaling pathways involved in cell growth, proliferation, metabolism, and apoptosis. Aberrant kinase activity is implicated in diverse human diseases, including cancer, neurodegeneration, and inflammatory disorders. Structure-based drug discovery, utilizing molecular docking and molecular dynamics (MD) simulations, has become a central strategy for identifying and optimizing STK inhibitors. In this review, we summarize recent advances and challenges in applying these in silico approaches to STK drug discovery. We discuss the principles, performance, and limitations of docking and MD approaches, as well as their integration with binding free-energy estimation methods. We emphasize recent methodological progress, including automated MD workflows, machine learning-driven interaction fingerprinting frameworks, and the growing adoption of hybrid docking-MD pipelines that enhance throughput and reproducibility. The review also highlights emerging directions such as computational design of heterobifunctional degraders (PROTACs) and allosteric modulators, which extend the scope of kinase targeting beyond ATP-competitive inhibitors. Quantitative examples of computational resource requirements and hit-validation rates from representative studies are summarized to contextualize the predictive power and practical feasibility of these approaches. Together, these developments demonstrate how the synergy of physics-based simulations, enhanced sampling, and machine learning is transforming MD from a purely descriptive technique into a scalable, quantitative component of modern kinase drug discovery. KEYWORDS

molecular docking, molecular dynamics simulations, serine/threonine kinases, drug discovery, STK inhibitors Frontiers in Pharmacology 01 frontiersin.org Hasan et al. 10.3389/fphar.2025.1696204 1 Introduction

There have been many successes, but kinase drug discovery continues to face challenges (Cohen et al., 2021). Selectivity is the most significant challenge, as the ATP-binding site, the canonical target for the majority of inhibitors, is highly conserved across kinases, leading to off-target binding risk and dose-limiting toxicity (Ferguson and Gray, 2018). Resistance, especially in cancer, is another major limitation, with members in the kinase domain sometimes mutated such that they do not bind inhibitors as well, leading to relapse (Cohen et al., 2021). Additionally, the intrinsic conformational flexibility of kinases poses a challenge for inhibitor development because these enzymes can exist in many different and distinct states, for example, active versus inactive conformations or aspartate-phenylalanine-glycine (DFG)-in versus DFG-out states of the activation loop (Schwartz and Murray, 2011). The identification and targeting of allosteric binding sites away from the ATP pocket provide one solution, but this approach does require very highresolution structural information (Govindaraj et al., 2022). Although traditional kinomics, led by experimental highthroughput screening drug discovery pipelines, have yielded numerous leads, they readily incur high costs, are timeconsuming, and lack the diversity of the chemical space they can access (Pollastri, 2011). Within this context, computational methods have developed into complementary and more rapid alternatives to experimental strategies (Khan et al., 2025). In particular, molecular docking and molecular dynamics (MD) simulations have become essential resources in kinase-targeted drug discovery (Naqvi et al., 2018). Docking is primarily used to predict the binding poses of small molecules to kinases (or similar structures) and their binding affinities, facilitating the virtual screening of large chemical libraries and the rational design of structure-activity relationships (Sousa et al., 2006). In contrast, MD simulations move beyond static docking models and consider the time-resolved flexibility of kinases and their complexes (Pikkemaat et al., 2002). Loop motions, activation states, solvent effects, and resistanceassociated mutations that are poorly sampled in validated rigid docking models can also be explored (Shukla and Tripathi, 2020). Docking and MD have been particularly useful in the initial stages of drug discovery against serine/threonine kinases (Roy et al., 2020; Ali et al., 2024; Khan et al., 2025). Docking can rapidly predict plausible binding modes of ligands while MD can refine those binding modes, assess their stability, and calculate the binding free-energy computed (e.g., via MM-PBSA or free-energy perturbation) (Vilar et al., 2008). Overall, this integrated workflow addresses the challenges of STKs, including difficulties in targeting essentially conserved ATP pockets, predicting the effects of resistance mutations, and characterizing potential allosteric sites that may not be readily apparent from static crystal structures (Lu et al., 2020). Such computational approaches are also valuable in the study of infectious diseases, as underexplored bacterial STKs represent promising targets for anti-virulence strategies and antibiotic-adjuvant therapies (Li et al., 2022). In this respect, the current review exemplifies the role of molecular docking and MD simulations as a discovery tool in the search for drugs against STKs. Here, we begin with an account of the structural and functional characteristics of the STKs, before proceeding to the specifics of the docking techniques and MD simulations, and how they can be integrated into drug discovery pipelines. We then discussed the main unresolved challenges,

Protein kinases represent one of the most extensive and most biologically important enzyme families in the human genome (Koch and Bajorath, 2025). They exert their regulatory functions in various cellular processes, including proliferation, differentiation, apoptosis, metabolism, and responses to environmental stress, by catalyzing the transfer of phosphate groups from ATP to the hydroxyl groups of specific amino acid residues in substrate proteins (Mencalha et al., 2014). Of these, serine/threonine kinases (STKs) constitute the most abundant class, accounting for over 70% of the kinome (Johnson et al., 2023). STKs act as molecular switches that fine-tune signaling cascades to regulate cell fate (Jin and Pawson, 2012). STKs are functionally important with well-known families, such as the mitogen-activated protein kinases (MAPKs), which mediate the effects of growth factors and cytokines (Moens et al., 2013); cyclin-dependent kinases (CDKs), which control cell-cycle progression (Malumbres et al., 2009); Akt and the mammalian target of rapamycin (mTOR), which integrate nutrient and energy signals affecting survival and growth (Castedo et al., 2002); AMP-activated protein kinase (AMPK), which acts as a metabolic sensor for restoring energy homeostasis (Sharma et al., 2023); and glycogen synthase kinase-3β (GSK3β) or cyclindependent kinase 5 (Cdk5), which have central roles in neuronal physiology and in neurodegenerative diseases (Yu H. et al., 2023). This broad functional repertoire underlines why STKs are frequently dysregulated in diverse pathologies, including cancer (Maoz et al., 2019), metabolic disorders (Rawat et al., 2023), and neurodegenerative diseases (Kawahata and Fukunaga, 2023). The clinical relevance of STKs is not restricted to human biology. Certain pathogenic bacteria also harbor eukaryotic-like STKs that contribute to stress responses, virulence, and antibiotic tolerance, as seen in Klebsiella pneumoniae (Hu et al., 2021; O’Boyle et al., 2025). KpnK kinase of K. pneumoniae promotes oxidative stress resistance and beta-lactam susceptibility, and HipA homologues mediate ciprofloxacin tolerance via autophosphorylation mechanisms similar to E. coli HipA (Srinivasan et al., 2014). While kinase research often focuses on human targets, recent findings suggest that STKs function as dualfunction molecules, playing a central role in both human disease regulation and bacterial pathogenicity, thereby broadening their applicability from oncology and neurology to the fields of infectious disease and antimicrobial resistance (Li et al., 2022). The drug targetability of kinases has been further demonstrated by the impressive number of clinically successful kinase inhibitors (Attwood et al., 2021a). To date, the United States Food and Drug Administration (FDA) has approved over seventy small-molecule kinase inhibitors since 2001, with many now targeting STKs in addition to the more traditional tyrosine kinases (Ayala-Aguilera et al., 2022). Palbociclib and other CDK4/6 inhibitors, for example, are now standard treatments for breast cancer (Liu et al., 2018), and everolimus and temsirolimus, mTOR inhibitors, are used clinically in oncology and tuberous sclerosis complex (Palavra et al., 2017). The increasing number of kinase inhibitors that have entered the clinic with demonstrated efficacy or safety finds high translational relevance in STK research (Attwood et al., 2021a). It emphasizes the urgency for new approaches to overcome long-standing hurdles in STK drug discovery.

2 diabetes, and metabolic syndrome-related diseases (Cao et al., 2025). In addition, STKs also influence inflammation, cardiovascular signalling, and immune reactions, expanding their clinical relevance (Mazzaschi et al., 2021). The significance of STKs extends beyond human diseases and is equally intriguing in bacterial systems (O’Boyle et al., 2025). Eukaryotic-like STKs play roles in antibiotic resistance and virulence in some bacteria, including K. pneumoniae (Srinivasan et al., 2014). For example, KpnK modulates stress adaptation and increases β-lactam resistance, and a HipA homologue has been shown to confer a biphasic response to ciprofloxacin via autophosphorylation (Li et al., 2022). These kinases are potential new antimicrobial targets that may lead to the selective attenuation of virulence or even the potentiation of currently used and inefficient antibiotics by targeting bacterial STKs (Li et al., 2022). The action of STKs as both therapeutic in human cells and anti-virulence in pathogenic organisms positions them at the unique intersection of oncology, neurology, metabolism, and infectious disease. Although STKs are attractive drug targets, the high selectivity and potency of STK inhibitors pose a challenge in drug discovery. The primary challenge among these is the extreme conservation of their ATP-binding sites, which poses a challenge for designing molecules to selectively target closely related kinases without compromising their selectivity versus others (Serafim et al., 2022). As a consequence, there is often off-target toxicity due to this lack of selectivity. Another key problem is the development of resistance, especially in the field of oncology, where missense mutations in the kinase domain may decrease the affinity of inhibitors (Lu et al., 2020). These resistance mutations frequently target the gatekeeper residue, the activation loop, or the DFG motif, reshaping the kinase conformational landscape. The intrinsic plasticity of these enzymes is another complicating factor, as kinases can exist in several different conformations following ligand binding or phospho-acceptor binding events (Hudmon and Schulman, 2002). Such conformational plasticity not only makes inhibitor design challenging but also complicates the computational prediction of efficacy, as static crystal structures typically do not represent the entire breadth of kinase states. STKs are characterized by their bilobal catalytic architecture, as well as their ATP-binding cleft and hinge region (Hardie, 1999). Conserved motifs, including the glycine-rich loop, the DFG sequence, and the activation loop, control nucleotide binding and catalysis, while the hinge provides a key hydrogen-bonding platform for inhibitor recognition. The conservation across kinases is extensive, and many ATP-competitive inhibitors target the same hinge interactions, making selective targeting challenging. However, variable and transient regions such as cryptic, allosteric pockets on kinase surfaces also provide attractive opportunities for targeting specificity. As shown in Figure 1, these architectural features are prominent in CDK2 and include numerous conserved hinge contacts and possible allosteric opportunities adjacent to the ATP-binding site (PDB: 1HCK). Although allosteric sites provide windows for selectivity, these sites are rarely constitutive and are often difficult to identify without the use of sophisticated structural or computational techniques (Lu et al., 2014). When viewed collectively, STKs offer both significant opportunities and challenges to drug discovery. Although their centrality in disease biology validates them as excellent

including selective mutagenesis, conformational heterogeneity, and computational cost and scoring, followed by future perspectives on machine learning (ML)-augmented simulations, hybrid quantum mechanical methods, and experimental structural biology methods such as cryo-electron microscopy. Through integration of recent case studies with methodological advancements, this article aims to deliver a unified narrative of how computational approaches are transforming therapeutic discovery against STKs in human and microbiome-related systems.

2 Structural and therapeutic significance of serine/threonine kinases STKs occupy a central role in cellular signaling because they phosphorylate serine or threonine residues on substrate proteins, thereby regulating downstream pathways that govern proliferation, differentiation, apoptosis, stress responses, and metabolism (Johnson et al., 2023). STKs contain a highly conserved bilobal catalytic domain characteristic of the kinase superfamily (Hardie, 1999). The smaller N-terminal lobe is predominantly β-sheet, containing the glycine-rich loop that stabilizes ATP-binding and the highly conserved lysine responsible for interaction with the phosphate groups of ATP (Roskoski, 2010). The C-terminal lobe, which is mainly α-helical, is substantially larger than the N-terminal lobe and forms the peptide substrate-binding interface. Within this conserved fold, multiple motifs are essential for catalysis and are also hot spots for anti-protein kinase drug design. It contains the hinge region, which binds ATP by hydrogen bonds and is a common binding position for inhibitors. Conformational changes in the activation loop switch kinases on and off (Gizzio et al., 2022). They are the primary determinants of the general state of kinase conformation and control the orientation of the magnesium ion necessary for catalysis, as seen in the DFG motif (Kung and Jura, 2016). Finally, the catalytic lysine in the β3 strand and a conserved glutamate in the αC-helix together position ATP for phosphotransfer. Such structural signatures both mediate kinase function and underpin the development of rational inhibitors. STKs are pivotal nodes in signaling networks, and thus, they are involved in various human diseases (Capra et al., 2006). The aberrant signaling through kinases like CDKs, MAPKs, Akt, and mTOR in cancer is a major contributor to driving uncontrolled proliferation, genomic instability, angiogenesis, and evasion of apoptosis (Stefani et al., 2021). As is well-known, CDK4/ 6 inhibitors like palbociclib have changed the treatment landscape for hormone receptor-positive breast cancer (Liu et al., 2018). In contrast, mTOR inhibitors such as everolimus have approvals in breast cancer, renal cell carcinoma, and tuberous sclerosis (Palavra et al., 2017). The MAPK pathway kinases, particularly the ERK subfamily, remain among the most extensively studied targets in the field of oncology (Braicu et al., 2019). In tauopathies, non-receptor kinases such as GSK3β and Cdk5 play crucial roles in tau hyperphosphorylation, synaptic failure, and neuronal demise, making them attractive therapeutic targets for Alzheimer’s disease, Parkinson’s disease, and related disorders (Yu H. et al., 2023). AMPK is a cellular energy sensor that modulates ATP levels by inducing catabolic pathways, making it a well-studied therapeutic target for the treatment of obesity, type

Structural architecture of serine/threonine kinases (STKs). (A) Representative STK catalytic domain (Cyclin-dependent kinase 2 (CDK2), PDB 1HCK) showing the conserved N-lobe (cyan), C-lobe (light orange), ATP-binding cleft and hinge (yellow), glycine-rich loop (purple), DFG motif (dark green), and activation loop (green). The bound ligand (magenta) illustrates canonical ATP-site engagement. (B) Close-up of hinge interactions highlighting the characteristic hydrogen-bond network that mediates broad ATP-competitive inhibitor binding. (C) Surface view of CDK2 reveals potential allosteric regions adjacent to the ATP pocket, highlighting the existence of cryptic binding sites that can be exploited for selectivity beyond the highly conserved ATP cleft. Structures were generated through PyMOL (DeLano, 2002) from the Protein Data Bank (Burley et al., 2022) entry 1HCK.

TABLE 1 Major families of protein serine/threonine kinases (STKs), representative members, their biological functions, and associated disease relevance. STK family Representative kinases Biological role

Disease relevance/ Therapeutic area Remarks/Inhibitor examples AGC family PKA, PKB/Akt, PKC, mTOR Cell survival, metabolism, and growth signaling Cancer, metabolic disorders, tuberous sclerosis Everolimus, Temsirolimus (mTOR); Perifosine (Akt)

CAMK family CaMKII, AMPK, DAPK Calcium signaling, energy sensing, and apoptosis regulation Neurodegeneration, type 2 diabetes, stroke Metformin (indirect AMPK activator); experimental DAPK inhibitors CMGC family

CDKs, MAPKs, GSK3, CLK Cell cycle control, stress response, neuronal regulation Cancer, inflammation, Alzheimer’s, Parkinson’s disease Palbociclib (CDK4/6), Trametinib (MAPK/ MEK), Tideglusib (GSK3β, experimental)

STE family MAPKKK kinases Regulation of MAPK cascades Cancer, immune, and inflammatory signaling Indirectly targeted via MAPK/ERK pathway inhibitors TKL family MLK, MLKL Developmental pathways, necroptosis

Inflammatory diseases, neurodegeneration Necrostatin-1 (RIPK1 inhibitor, experimental) RGC family Guanylate cyclase kinases cGMP-dependent signal transduction Cardiovascular disease, metabolic disorders

Few selective inhibitors; potential in vascular biology Bacterial STKs HipA, KpnK (K. pneumoniae) Stress response, virulence, and antibiotic tolerance Antimicrobial target; drug resistance modulation Novel target class; inhibitors under preclinical exploration

Approved and investigational inhibitors are also highlighted, emphasizing the broad therapeutic spectrum of STKs, in oncology, neurology, metabolism, and infectious disease.

take advantage of computational approaches, including molecular docking and MD simulations, are increasingly bridging this gap by providing insights of kinase specificity, conformational flexibility,

therapeutic targets, the structural conservation of their catalytic domains, their conformational heterogeneity, and their propensity for resistance mutations will require novel strategies. Methods that

Frontiers in Pharmacology 04 frontiersin.org Hasan et al. 10.3389/fphar.2025.1696204 the next steps in medicinal chemistry. A further obvious application is selectivity profiling, where candidate inhibitors are docked against panels of related kinases (Zhong and Almahmoud, 2023). Selectivity is a major challenge in kinase drug design, as the ATP-binding site is highly conserved across the kinome, making docking-based profiling a valuable first step towards predicting off-target interactions that can be subsequently tested experimentally. In the case of specifically in STKs, docking has been used to develop inhibitors that capitalize on minor variations in shape and electrostatics of the binding pocket. Docking analyses have guided the identification of compounds that specifically bind CDK4/6 relative to other CDK isoforms and have also revealed key interactions in the hydrophobic pocket next to the hinge region in mTOR inhibitors (Najmi, 2025). Docking is also increasingly applied to drug repurposing, where existing FDAapproved drugs are screened against STK targets to find new possible therapeutic uses. This strategy is particularly attractive, as the pharmacokinetic and safety profiles of repurposed drugs have already been determined, allowing for a more rapid translation to the clinic.

and inhibitor optimization that cannot be easily achieved through experiments (Naqvi et al., 2018). To highlight the richness and potential therapeutic relevance of STKs, Table 1 summarizes the major families of STKs, their representative members, biological functions, and provides examples of both approved and investigational inhibitors.

3 Molecular docking approaches in STK inhibitor discovery 3.1 Principles of docking Molecular docking is a structure-based computational method that predicts the binding mode and affinity of small molecules in the active site or allosteric site of proteins (Vilar et al., 2008). In the case of kinases and especially STKs, most docking studies have concentrated on the ATP-binding pocket, which is the most conserved and pharmaceutically targetable site in kinases (Ikram et al., 2019). Docking involves two main steps: (i) sampling, which generates possible ligand poses, and (ii) scoring, which evaluates these poses using scoring functions (Trott and Olson, 2010). In general, sampling algorithms aim to consider all possible orientations and/or conformations of a ligand with respect to the protein binding site. In contrast to rigid docking, flexible docking allows for partial rearrangements of side chains or backbone elements, simulating the features of the induced fit (Mohanty and Mohanty, 2023). However, this inherent flexibility of kinases could be dealt with better in advanced ensemble docking approaches that include multiple receptor conformations, typically obtained from crystallographic or MD simulation studies. Scoring functions estimate binding affinities to rank binding poses generated from docking calculations (Hassan et al., 2017). They can be empirical, knowledge-based, or derived from molecular mechanics force fields, and evaluate the contributions of hydrogen bonding, hydrophobic bulk interactions, electrostatics, and van der Waals packing (Trott and Olson, 2010). While scoring functions are useful, they are approximate and may not accurately recapitulate experimental binding energies. To address these limitations, consensus scoring (combining multiple scoring functions) or post-docking refinement (using MM-PBSA calculations, for instance) is widely employed (Wang et al., 2019).

3.3 Docking success stories in kinase drug discovery Several landmark examples highlight the importance of docking in drug discovery for kinases (Attwood et al., 2021a). Imatinib, such a targeted agent, is actually a pan-tyrosine kinase inhibitor and serves as a paradigm for future STK inhibitor development (Di Vito et al., 2023). BCR-ABL is in the autophosphorylated state, and docking studies have shown that imatinib stabilizes the inactive conformation by forming hydrogen bonds with the hinge region and binding in the hydrophobic back pocket exposed in the DFG-out state (Rocha et al., 2021). The success of this was translated to STKs, where inhibitors were similarly optimized to exploit conformational states. For example, docking-guided structure-activity relationship studies were instrumental in identifying and optimizing hingebinding motifs that imparted isoform selectivity in the case of CDK inhibitors, such as palbociclib and ribociclib (Braal et al., 2021). In recent years, however, docking-based drug repurposing has found surprising interactions of approved drugs on STKs (Wang et al., 2024). Recently, antidiabetic drugs that activate AMPK have been repurposed, and several anticancer agents have been experimentally validated as mTOR inhibitors (Khan et al., 2024). Such success stories highlight both the power (to generate structural hypotheses) and the weaknesses of docking. Docking predictions were often refined with MD simulations and/or validated using crystallography and biochemical assays (Huang and Hu, 2025). However, docking remains the initial step in the computational pipeline for discovering kinase inhibitors, providing a rapid screen of vast chemical spaces, insight into binding interactions, and aiding in the rational design of more potent and selective inhibitors. Several software platforms are available for performing docking studies of kinases, each with its own merits and demerits, and therefore preferred for specific applications. An overview of the docking programs frequently used for various types of proteins, including STKs, is given in Table 2.

3.2 Docking applications in STKs Docking plays a pivotal role at multiple stages of STK inhibitor discovery (Zhong et al., 2022). Virtual screening is one of the most common applications, where thousands of chemical libraries are docked into the binding pocket of a kinase to identify useful scaffold hits (Mohammad et al., 2020b). Such a strategy minimizes the number of candidates to be validated experimentally, saving time and cost (Alrouji et al., 2025). Docking also supports binding mode prediction, enabling the visualization of inhibitor binding to important kinase motifs, including the hinge region, the conserved catalytic lysine, or the DFG motif. Such information is useful for understanding structure-activity relationships and aiding

TABLE 2 Widely used molecular docking software platforms for kinase inhibitor discovery, including open-source and commercial tools. Software Type Sampling method Scoring function Strengths Limitations

Applications in kinase studies InstaDock Open-access (GUI for QuickVina-W) Flexible ligand, semi-rigid receptor Vina scoring User-friendly GUI; batch screening; accessible to non-programmers Limited receptor flexibility; less customizable

Virtual screening of large libraries; kinase-focused repurposing screens AutoDock/ AutoDock Vina Open-source Lamarckian genetic algorithm (AutoDock); gradient optimization (Vina) Empirical free energy scoring

Widely used; flexible ligand; semi-rigid receptor; good community support Scoring function relatively simple; limited allosteric handling Broad kinase inhibitor screening; hinge-binding motif analysis DOCK

Open-source Grid-based matching Force-field based Early and efficient tool; handles large libraries well Older interface; less advanced handling of protein flexibility Used in early MAPK and CDK docking campaigns

Glide (Schrödinger) Commercial Systematic search with grid-based potentials GlideScore High accuracy; multiple precision modes (HTVS, SP, XP) Proprietary; requires license; high cost Benchmark kinase inhibitor design; hinge region SAR optimization

GOLD Commercial Genetic algorithm ChemScore, ASP, GoldScore Robust handling of ligand flexibility; reliable for kinases Proprietary; performance depends on the scoring function Selectivity profiling across kinase families (e.g., CDKs, MAPKs)

CDOCKER (Discovery Studio) Commercial CHARMm-based MD docking Force-field based Explicit receptor flexibility; MD refinement of docking Limited to the Discovery Studio platform; license required Applied to mTOR and CDK inhibitor optimization

RosettaLigand Open-source Monte Carlo + minimization Rosetta energy function Good induced-fit handling; flexible docking Complex workflow; steeper learning curve Allosteric site exploration in STKs; flexible loop docking

Each entry summarizes the sampling method, scoring function, strengths, and limitations, with representative applications in serine/threonine kinase (STK) research.

capture the full dynamic range of kinases. STKs are highly flexible enzymes, like other members of their superfamily; the transition between different conformations is an essential part of their function (O’Boyle et al., 2025). This includes changes in the conformation of the activation loop, the glycine-rich P-loop, and the DFG motif, which can result in rapid and sometimes large alterations of ligand accessibility to the binding pocket and/or the binding affinity of the ligand to the target (Schwartz and Murray, 2011). MD simulations overcome these limitations by solving Newton’s equations of motion for systems of atoms and by offering time-resolved, atomic-time trajectories of protein-ligand complexes (Fu et al., 2022). MD simulations allow exploration of broader aspects of protein flexibility, solvation, ion coordination, and inter-residue watermediated interactions that are seldom present during docking studies. MD allows the user to observe how a kinase toggles between these states, how an inhibitor stabilizes or destabilizes those states, and whether water molecules play a role in essential hydrogen bonding networks in the binding site. Crucially, MD tests docking-derived poses for stability under physiological conditions, which ensures that such interpretations of binding modes are not merely artifacts of rigid docking algorithms.

3.4 Choosing docking strategies for orthosteric vs. allosteric/cryptic pockets For STKs, the ATP (orthosteric) site is well-defined and generally well-handled by grid-based and standard flexible ligand docking approaches that assume limited receptor rearrangement. Tools such as AutoDock Vina and Glide (HTVS/SP) are efficient for large-scale orthosteric virtual screening and hinge-motif SAR exploration (Trott and Olson, 2010). In contrast, allosteric and cryptic pockets typically require explicit receptor flexibility or ensemble approaches. Methods such as induced-fit methods (e.g., RosettaLigand, GOLD with flexible sidechains, Glide Induced-Fit), MD-derived ensemble docking, or MD-refined docking (e.g., CDOCKER with MD refinement) are more suitable (DeLuca et al., 2015; Wu and Brooks III, 2021). For cryptic pockets that open transiently, generating receptor conformations by enhanced sampling MD (metadynamics, GaMD, replica-exchange) or by short, targeted MD, then using ensemble docking across those conformations is recommended (Kuzmanic et al., 2020b). Finally, consensus and rescoring strategies, e.g., docking, short MD, MMGBSA rescoring, often perform best when seeking selective allosteric modulators.

4 Molecular dynamics simulations in STK inhibitor design

There has been a growing application of MD to STK drug discovery in recent years, and multiple different roles have emerged (Attwood et al., 2021b). One important application is docking pose validation (Alzain et al., 2025). MD simulations in an explicit solvent can also be used to relax the protein-ligand complex and explore whether the interactions remain stable over nanosecond to microsecond time scales after a successful docking experiment,

4.1 Fundamental role of MD in kinase studies Molecular docking provides a quick perception of potential ligand binding orientations within protein active sites. However, docking assumes a relatively static protein structure and fails to

Frontiers in Pharmacology 06 frontiersin.org Hasan et al. 10.3389/fphar.2025.1696204 alchemical methods, such as free energy perturbation (FEP) and thermodynamic integration, yield higher accuracy but also at a significantly larger computational expense (Ruiz-Blanco and Sanchez-Garcia, 2020). In practice, MM-PBSA and MM-GBSA are applied as end-point estimators on snapshots extracted from production MD trajectories (Genheden and Ryde, 2015). Typical workflows perform energy decomposition to parse contributions from van der Waals, electrostatic, polar solvation (using PB or GB), and nonpolar solvation. These methods are computationally inexpensive relative to alchemical FEP/TI and are therefore widely used to rerank docking hits and to prioritize analogues for synthesis. However, MM-PBSA/MM-GBSA accuracy depends strongly on sampling quality, choice of dielectric and surface tension parameters, and force-field consistency between MD and ligand parameterization. For kinases, where solvent networks and flexible loops can substantially influence binding energetics, it is advisable to extract energies from multiple independent replicate simulations to quantify statistical uncertainty, to report the mean and standard deviation of the calculated ΔG values, and to validate MM-PBSA/ MM-GBSA results against at least a subset of experimental affinities before relying on them for decision making. When higher accuracy is required during lead optimization, alchemical free-energy methods such as FEP or TI remain the benchmark approaches despite their greater computational cost. Other notable advances are enhanced sampling techniques (Lazim et al., 2020). Kinetic traps frequently constrain standard MD, as proteins can reside entrapped in local conformations that may not reflect the complete conformational landscape (Kuzmanic et al., 2020a). Now, we have methods such as accelerated MD, metadynamics, replica-exchange MD, and Gaussian accelerated MD to bypass these barriers, unveiling hidden conformations and improving conformational sampling (Wang et al., 2021). The advantages of these methods have previously helped dissect activation loop dynamics, pinpoint cryptic allosteric sites, and study conformational selection during ligand binding in kinases (Kuzmanic et al., 2020a). Lastly, the trends of MD with structural biology and artificial intelligence (AI) are future directions in kinase studies (Agajanian et al., 2023). Importantly, the last few years have seen tangible improvements in throughput, automation, and downstream analysis of MD-based hit refinement. Automated MD pipelines that streamline setup, execution, and post-processing of many protein-ligand simulations now exist and have been applied to accelerate hit prioritization (Brueckner et al., 2024). Examples include Admiral, an automated docking, MD, and analysis platform that orchestrates simulation setup, runs, and automated reporting for medicinal chemistry teams, and recent automated MD workflows that integrate ML models to generate per-ligand simulation fingerprints and prioritize candidates (Baumgartner and Zhang, 2020). Complementary to automation, tools for encoding molecular interactions from MD trajectories as compact fingerprints have facilitated rapid comparisons and ML-driven analyses. Libraries such as ProLIF enable the extraction of interaction fingerprints from trajectories and trajectory-derived ensembles, allowing clustering of ligand binding modes, feature engineering for ML models, and rapid filtering of MD-derived poses (Bouysset and Fiorucci, 2021).

suggesting potential inhibitors (Roy et al., 2020). Stable trajectories imply real predictions of docking, rapid dissociation of the ligand, or significant rearrangements of the complex indicate a false positive. MD is also fundamental to the crystallographic analysis of the conformational flexibility of STKs (Gizzio et al., 2022). Kinases frequently toggle between DFG-in and DFG-out configurations, as well as open and closed states of the activation loop or inward- and outward-facing conformations of the αC-helix. They help determine whether inhibitors can bind to active or inactive conformations, and as such are crucial to the design of inhibitors. Simulations illustrated mechanisms at an atomic level, explaining how the inhibitors bias protein kinases to use the inactive conformation over the active conformation. A third key application is the investigation of resistance mutations (Yu Y. et al., 2023). Mutations that change the conformational dynamics or steric environment of the binding pocket often leads to clinical resistance. MD simulations have been utilized to model these mutations, indicating changes in hydrogen bonding networks, disruptions in hydrophobic packing, and alterations in inhibitor-bound state stability (Mohammad et al., 2020a). While the extensive literature on resistance mechanisms has targeted tyrosine kinases, such as EGFR or BCR-ABL, the same paradigms are relevant to STKs, as resistance mutations can limit the clinical utility of CDK or mTOR inhibitors (Alves et al., 2021). MD may also be one of the most valuable tools for the discovery of allostery (Govindaraj et al., 2022). In contrast to ATP-competitive inhibitors that target the conserved catalytic pocket, allosteric inhibitors utilize noncatalytic, often transient sites. Such sites are hard to discern with static crystallography but are well exposed by long MD simulations that can reveal opening and closing motions or expose cryptic pockets. Simulations of mTOR have, for instance, revealed hydrophobic pockets that lie outside the canonical ATP-binding site, which are currently being explored for their potential as allosteric regulators (Nunes Azevedo et al., 2023).

4.3 Recent advances Recent methodological and computational advances have greatly improved the utility of MD for targeted multi-scale drug discovery against kinases (Sadybekov and Katritch, 2023). Meanwhile, GPU acceleration, or the availability of specialized hardware (such as Anton supercomputers), allowed the extension of the simulation time window from nanoseconds to microseconds and even milliseconds (Shaw et al., 2021). Such extended simulations enhance conformational sampling and capture rare yet biologically relevant transitions, such as activation loop unfolding or ligand unbinding events. Moreover, this qualitative understanding of ligand binding has been coupled with several binding free-energy methods on MD, and this has allowed for an increasingly quantitative prediction of inhibitor affinity. Molecular Mechanics/Poisson-Boltzmann Surface Area (MM-PBSA) and Molecular Mechanics/Generalized Born Surface Area (MMGBSA) are post-processing approaches that facilitate fast, albeit approximate, binding free energy calculations from MD trajectories (Wang et al., 2019). Therefore, more rigorous

TABLE 3 Major molecular dynamics (MD) software packages employed in kinase simulations. Software License Strengths Limitations Applications in kinase research GROMACS Open-source Swift; strong GPU acceleration; widely used in academic labs; large community support

Limited force-field variety compared to AMBER/CHARMM; less intuitive for absolute binding free-energy methods

Validation of docking poses; long-timescale simulations of MAPKs, CDKs, and mTOR; widely used for MM-PBSA in kinase-ligand studies AMBER Commercial/ academic licenses

Rich library of biomolecular force fields (ff14SB, GAFF); strong MM-PBSA/MMGBSA support; good integration with quantum mechanics (QM/MM) tools Slower than GROMACS for extensive systems; license restrictions for some components

Free-energy calculations for CDKs and Akt inhibitors; QM/MM studies of catalytic residues in STKs NAMD Open-source Highly scalable on large clusters; efficient CHARMM force field support; good for extensive systems

Moderate learning curve; less userfriendly for beginners

Long-timescale simulations of kinase conformational changes (e.g., DFG-in/out transitions); ensemble simulations for inhibitor selectivity CHARMM Commercial (academic version available)

Highly detailed biomolecular modeling; extensive force-field options; strong for advanced free-energy methods Complex input and steep learning curve; less streamlined than GROMACS/ AMBER Detailed mechanistic studies of ATP binding in kinases; conformational plasticity analysis of STKs

Desmond (Schrödinger) Commercial Extremely fast; optimized for GPUs; seamless integration with Glide docking results Proprietary; limited customizability compared to open-source tools

Kinase inhibitor optimization pipelines (Glide docking, Desmond MD refinement); mTOR and CDK inhibitor refinement OpenMM Open-source

Highly flexible and customizable; strong GPU acceleration; Python-based API makes integration with ML easy Still in development; has a smaller user base; fewer validated workflows than GROMACS/AMBER.

AI-driven kinase simulations; adaptive sampling of STK conformations; emerging tool for integration with ML-enhanced workflows

The table outlines license type, strengths, limitations, and representative applications in studying serine/threonine kinase (STK) structure, conformational flexibility, and inhibitor binding.

candidates (Huang and Hu, 2025). This is especially useful for STKs for which the experimental high-throughput screening is expensive and inefficient due to the similarity of the conserved ATP-binding site (Zhang et al., 2022). Docking has the potential to identify ligands that utilize small differences in hinge regions, hydrophobic pockets, or allosteric cavities, and to create testable hypotheses about selectivity and potency, guiding downstream computational and experimental assays. Molecular docking and MD simulations constitute complementary approaches that, when combined, offer a highthroughput and evidence-driven pipeline for kinase-targeted drug discovery. Docking acts as a quick initial layer for virtual screening and pose prediction. In contrast, MD then refines these predictions in physiologically relevant environments, permitting the inclusion of protein flexibility, solvent effects, and dynamic stability. Together, they enable better predictive power for binding depth and more effective prioritization of candidate inhibitors. Figure 2 provides a schematic overview of this integrated workflow in the context of STK inhibitor discovery.

When combined with automated MD workflows and adaptive sampling, interaction fingerprinting supports scalable, reproducible post-processing of large MD datasets and enhances the interpretability of ML models trained on dynamic interaction patterns. MD simulations have become a staple in providing dynamic context to structures obtained from experimental methods such as cryo-electron microscopy, NMR, and X-ray crystallography (Son et al., 2024). This led to the employment of ML approaches that utilize large MD datasets to pull out essential collective variables and expedite the conformational sampling process (Wang et al., 2020). This, in turn, enhances the reach and precision of MD, making it a cornerstone in rational STK inhibitor discovery. There are various MD packages, each with specific pros and cons that limit their application to kinase simulations. Conventional MD engines and their applications in the discovery of STK inhibitors are summarized in Table 3.

5 Integrative docking-MD workflows in STK drug discovery 5.1 Docking as the first step 5.2 MD for refinement and validation

Rational drug discovery can greatly benefit from a stepwise integration of molecular docking and MD simulations, which are complementary approaches today (Sadybekov and Katritch, 2023). Docking is typically employed as an initial step due to its speed and generality in screening large compound libraries against kinase targets. The ability to effectively explore vast chemical spaces and identify possible ligand binding poses, ranking them based on scoring functions, allows docking to help researchers effectively narrow down chemical spaces to a reasonable subset of

MD simulations are used to identify, characterize, and validate promising compounds through docking under dynamic and physiologically relevant conditions after all compounds have been docked (Lazim et al., 2020). While docking usually considers the protein rigid, MD considers the conformational flexibility of both ligand and receptor, and also the solvent effects and long-range electrostatics. In the final stage, the stability of docking poses is evaluated using molecular dynamics simulations of the proteinligand complex in explicit solvent, spanning nanosecond to

Schematic overview of an integrated computational pipeline for serine/threonine kinase (STK) inhibitor discovery. Molecular docking (left) enables the identification of the binding pocket, virtual screening of chemical libraries, prediction of binding poses, scoring, and prioritization of hits. Molecular dynamics simulations (middle) refine docking predictions by evaluating ligand-kinase complex stability, conformational flexibility, solvent effects, and lead optimization. Integration of docking and MD (right) allows free energy calculations (e.g., MM-PBSA, FEP), estimation of binding affinities, and selection of stable candidate inhibitors for experimental validation. Together, these complementary approaches provide both breadth (docking-based exploration) and depth (MD-based refinement) in kinase-targeted drug discovery.

et al., 2016; Hassan et al., 2023). One recent study focusing on CDK1 initially used docking to screen commercially available databases of candidate inhibitors that were refined in ranking through MD simulations and MM-PBSA calculations (Teotia et al., 2024). Among the highest-ranked compounds, several were found to have micromolar inhibitory activity in vitro, thereby validating the predictions made from computational analysis. Docking was used to identify compounds that not only bind to ATP-competitive sites but also to allosteric sites in mTOR inhibitors; similar strategies have been applied (Dahiya et al., 2019; Gupta et al., 2019; Botelho et al., 2022). While MD simulations also confirmed the binding stability of these inhibitors, they revealed dynamic movements of the kinase domain that were not apparent from static docking results. An additional illustrative example is the case of salt-inducible kinases (SIKs), where ensemble docking using MD-derived conformations improved the correlation between the predicted docking score and the log of experimental IC50 values (ValdésAlbuernes et al., 2025). These case studies exemplify how docking brings breadth-rapid exploration of chemical space, while MD contributes depth, dynamic validation, and energetic optimization. This represents a rational, iterative framework for kinase inhibitor discovery encompassing both docking and MD. Docking creates the first hypotheses regarding binding poses and possible selectivity. At the same time, MD interrogates and refines these hypotheses, providing insight into conformational dynamics, resistance mutations, and solvent-exclusion-mediated interactions (Tesch et al., 2021). These integrated approaches significantly enhance the efficiency and fidelity of computational pipelines, enabling translation of in silico predictions to validated kinase inhibitors in the lab. Due to the importance of the STKs in cellular signalling, several recent studies combining docking with MD simulations have been performed that identified and optimized

microsecond timescales (Fu et al., 2022). If the docking-predicted pose remains stable, the inhibitor is more likely to be a true binder; conversely, ligand dissociation or major conformational rearrangements may indicate a false positive. MD further enables side chains in flexible kinase motifs, such as the activation loop, P-loop, or DFG motif, to relax and fit the binding ligand, providing more realistic perspectives on binding (Shukla and Tripathi, 2020). The other improvement step is to extract the binding free energies from MD trajectories. In addition to re-ranking docking hits with approximate approaches such as MM-PBSA or MM-GBSA, alchemical methods like FEP permit quantitative affinity predictions (Wang et al., 2019; Ruiz-Blanco and Sanchez-Garcia, 2020). This is useful in avoiding some of the biases of docking scoring functions, which are generally poorly or only moderately related to experimental binding affinities. The combined use of docked hits and MD-based free energy calculations enables the generation of a reliable ranking of potential inhibitors, which can effectively limit the number of compounds to be synthesized and tested experimentally.

5.3 Example workflows and case studies The effects of this docking and MD harmony are best seen with combined workflows (Shaikh et al., 2023). The standard pipeline starts with docking large libraries of compounds against an STK target in virtual screening (Al-Fahad et al., 2025). This step reveals stable ligands that then undergo MD simulations to confirm their relative stability in the binding pocket. These simulations yield binding free energies that are used for ranking, and then the most promising candidates are chosen for experimental testing (Zhang et al., 2024). Such a two-pronged strategy has proven successful in discovering inhibitors for multiple STKs (Tarazi

TABLE 4 Representative case studies (2020–2025) of serine/threonine kinase (STK) inhibitor discovery using integrated docking and molecular dynamics (MD) approaches. Target kinase Computational approach

#Compounds screened (library/source) Experimentally validated hits/ Hit rate (%) Key outcome/ Findings Therapeutic context Year