Blind molecular docking, interaction mapping, and ADMET-based prioritization were combined in this in silico original research study. The procedure created a 3 × 3 descriptive docking matrix comparing three ligands across three molecular targets associated with H. pylori. The analysis was not intended to replace biochemical or microbiological validation; it was a hypothesis-generating screening phase.

Shikimate kinase, DNA gyrase subunit A, and H. pylori urease were selected to represent complementary biological functions: an essential microbial biosynthetic pathway, DNA topology, and acid adaptation and colonization, respectively. Protein Data Bank (PDB) 1E9Y was chosen for urease because of its established biological relevance and structurally defined nickel-containing architecture. PDB 1ZUH was used for shikimate kinase because it is an experimentally determined H. pylori structure. DNA gyrase subunit A was represented by the AlphaFold model linked to UniProt P48370 because a suitable experimental structure for the planned blind-docking comparison was unavailable. Biological relevance does not by itself establish druggability; accordingly, these proteins were treated as plausible test targets, and tractability was not assumed from docking alone.

Metal components important for urease stability were retained, non-essential water molecules were removed, and protein files were visually examined. This step was performed to avoid disruption of the nickel-dependent architecture of urease.

Levofloxacin, berberine, and niclosamide were the ligands under investigation. Levofloxacin was used as a comparator with established clinical significance in H. pylori rescue therapy, berberine as a naturally occurring bioactive alkaloid, and niclosamide as a repurposing candidate. Three-dimensional ligand structures were created or acquired in CB-Dock2-compatible formats and examined for structural consistency before submission. Berberine was represented in its permanent quaternary ammonium form. Alternative protonation, tautomeric, and ionization microstates were not systematically enumerated for niclosamide or levofloxacin; this preparation constraint is acknowledged because ligand state can influence predicted poses and scores.

Accordingly, the ligand panel was guided by the literature. Levofloxacin was retained only as a clinically recognized antibacterial comparator, while niclosamide and berberine were given priority because of their prior anti-H. pylori-related data. The goal of the docking experiment was not to replicate the entire clinical pharmacology of eradication therapy but rather to investigate possible enzyme-level engagement.

Because levofloxacin has been linked to bacterial topoisomerase targeting and is used in rescue anti-H. pylori regimens, it was included as a therapeutically relevant comparator. It was not used to determine the clinical superiority of the repurposed compounds and was not considered a novel candidate. Since the goal was to compare two non-classical structures with a clinically well-known antibacterial reference rather than to replicate all therapeutic classes, clarithromycin and metronidazole were excluded from the docking matrix.

All ligands were assessed across the same target set, residue-level interaction maps were created, and the best-ranked poses were visually examined in three dimensions. These consistency checks aid interpretation but do not constitute formal validation of the docking protocol. No cognate-ligand redocking, receiver operating characteristic benchmarking against known binders and nonbinders, or prospective experimental validation was performed; therefore, pose accuracy and enrichment cannot be claimed. Docking was performed using CB-Dock2, an automated blind-docking server that combines cavity detection, docking, and pose ranking (Trott and Olson 2010; Forli et al. 2016; Liu et al. 2022). For each receptor-ligand pair, the top-ranked pose according to the Vina score was retained. Vina-type outputs were interpreted as relative docking scores, not experimental binding constants or binding free energies. Discovery Studio Visualizer was used only to inspect three-dimensional poses and residue-level interaction maps. The workflow is therefore suitable for hypothesis generation, not for demonstrating enzyme inhibition or antibacterial activity.

Protein structures were identified by the PDB and UniProt identifiers reported above, and ligand structures were obtained from PubChem. CB-Dock2 (Liu et al. 2022), PubChem, and SwissADME were accessed in June 2026. The server-side CB-Dock2 build number was not displayed to the user. Discovery Studio Visualizer (BIOVIA, Dassault Systèmes) was used for three-dimensional inspection and two-dimensional interaction diagrams; the locally installed build metadata were not retained. Original docking outputs, interaction files, and figure panels were retained. These disclosures identify the available reproducibility information without implying unavailable version precision.

Vina-type outputs were retained in their native scoring units and were not converted to Ki or pKi. A docking score is an empirical ranking function and is not the binding free energy, delta G binding = G complex – (G free protein + G free ligand). Docking does not determine all free-state and solvent contributions required by this thermodynamic relationship; a favorable score can therefore occur for a nonbinder. Because the matrix contained only three ligands and three targets, no inferential statistical test was used. Means and standard deviations are reported only as descriptive summaries of this specific matrix and cannot support population-level or biological inference.

Molecular weight, topological polar surface area, lipophilicity, hydrogen-bond donors and acceptors, predicted gastrointestinal absorption, blood–brain barrier permeation, Lipinski rule-of-five compliance, bioavailability score, and medicinal-chemistry alerts were obtained from SwissADME-type workflows (Lipinski et al. 2001; Veber et al. 2002; Daina et al. 2017). These are computational descriptors, not measurements of exposure, bioavailability, safety, or therapeutic suitability. They were used to flag formulation and pharmacokinetic questions rather than to validate or upgrade the docking ranking.

The chosen best-ranked poses were reorganized into target-specific two-dimensional interaction maps to supplement the numerical docking matrix, the representative three-dimensional binding poses, and the residue summary in Table 3. This format makes it possible to compare the interaction patterns of levofloxacin, berberine, and niclosamide within the same receptor context. The two-dimensional diagrams served as visual support for the plausibility and chemical interpretation of the chosen docking poses rather than as a separate scoring mechanism. The corresponding urease, DNA gyrase A, and shikimate kinase maps are presented in Figs 4, 5, 6, respectively.

Overall, the interaction maps show that poses with the same Vina score may contain different predicted contact patterns. This was especially noticeable for urease, where berberine and niclosamide had different dominant interaction classes despite identical scores. The maps are descriptive visualizations of predicted poses and were not treated as independent evidence of binding or biological activity (Table 4).

Urease: All three top-ranked poses occupied the same predicted cavity region. Niclosamide and berberine produced the same score (–8.5 kcal/mol), but this numerical equality does not demonstrate equal affinity or inhibition. Berberine’s predicted pose contained mainly pi-associated and hydrophobic contacts, whereas the niclosamide pose contained hydrogen-bonding, hydrophobic, and aromatic contacts. Levofloxacin produced a score of –7.9 kcal/mol with a broad predicted residue-contact pattern.

DNA gyrase subunit A: Berberine produced the most negative score in this three-ligand comparison (–8.5 kcal/mol), with a predicted pose containing aromatic contacts involving PHE A:549 and TYR A:525. Niclosamide scored –7.8 kcal/mol and levofloxacin –7.5 kcal/mol; these differences do not establish affinity or potency. Interpretation is further limited because the receptor was not an experimentally resolved cocrystallized DNA gyrase–drug complex, and fluoroquinolone action in vivo depends on the ternary DNA–enzyme–drug context.

Shikimate kinase: Scores against shikimate kinase were less negative than those for urease and DNA gyrase A in this matrix. Berberine ranked first (–7.3 kcal/mol), followed by levofloxacin (–7.0 kcal/mol) and niclosamide (–6.9 kcal/mol), but the small differences should not be interpreted as potency differences. Shikimate kinase remains conceptually relevant because the classical shikimate biosynthetic pathway is present in many microorganisms and plants but is not encoded as a corresponding pathway in humans, making its enzymes attractive, though not automatically tractable, antimicrobial targets.

The current computational study should not be interpreted as suggesting an antibiotic substitute but rather in the context of guideline-directed eradication therapy. A resistance-aware strategy is supported by the Maastricht VI/Florence Consensus Report, the American College of Gastroenterology (ACG) guideline, and the American Gastroenterological Association (AGA) Clinical Practice Update. Local resistance, prior antibiotic exposure, susceptibility testing when available, appropriate duration, and adherence should all be taken into account when choosing a regimen (Shah et al. 2021; Malfertheiner et al. 2022; Chey et al. 2024). Docking/ADMET research can be useful within this framework if it is presented as a prioritization step that identifies compounds for further biochemical and microbiological confirmation.

Existing experimental evidence shows that niclosamide can inhibit H. pylori growth under the tested conditions, retain activity at acidic pH, affect adhesion/invasion, and show no readily detectable resistance during serial exposure experiments (Tharmalingam et al. 2018). The present docking results neither reproduce those findings nor identify their mechanism. They generate a separate, target-oriented hypothesis that urease and the other selected enzymes could be tested experimentally; the predicted poses cannot establish that these proteins mediate the previously observed antibacterial effects.

Meta-analytic and clinical data indicate that berberine-containing combination regimens may improve or achieve clinically relevant eradication outcomes in some settings (Zhang et al. 2017; Hu et al. 2020), whereas other results do not demonstrate clear superiority (Chen et al. 2023). These clinical observations cannot validate the enzyme poses predicted here, and the docking results cannot identify the mechanism responsible for reported clinical or antibacterial effects. Berberine should therefore be viewed as a scaffold for mechanistic and formulation-focused testing, not as a docking-supported monotherapy.

The inclusion of urease, DNA gyrase A, and shikimate kinase was intended to cover complementary biological levels: acid adaptation and colonization, DNA replication/topology, and the classical shikimate biosynthetic pathway, which is not encoded as a corresponding pathway in humans. Recent anti-H. pylori work using natural compounds and in silico enzyme targeting supports this type of multitarget framework when computational findings are subsequently connected to biochemical or microbiological validation (Noori Goodarzi et al. 2024). In the present study, the defensible interpretation is candidate prioritization rather than therapeutic prediction.

The most pharmacologically intuitive results are those for urease. All three ligands occupied a predicted urease cavity with favorable or moderate scores. Urease is a nickel-dependent virulence enzyme that is necessary for adaptation to the gastric niche. Although berberine and niclosamide had the same urease score, their interaction maps revealed distinct stabilization patterns. This observation demonstrates that equal Vina scores should not be interpreted as mechanistically identical binding modes, which supports the use of two-dimensional Discovery Studio Visualizer diagrams in the main article.

The results for DNA gyrase A need to be interpreted more cautiously. Although levofloxacin is clinically associated with bacterial topoisomerase inhibition, its actual mechanism involves the enzyme–DNA complex and quinolone resistance-determining regions. Thus, the pharmacodynamic context of fluoroquinolone action cannot be properly replicated by the isolated DNA gyrase A model. However, berberine demonstrated a favorable predicted interaction driven by hydrophobic and aromatic contacts, which is in line with the increased interest in anti-H. pylori candidates derived from berberine (Tharmalingam et al. 2018).

The shikimate kinase scores were less negative than those for the other targets in this matrix. The enzyme belongs to the classical shikimate pathway, which is present in many microorganisms and plants but is not encoded as a corresponding pathway in humans; this biological distinction motivates testing but does not establish tractability. The small score differences do not show that any compound inhibits shikimate kinase. For the present ligands, this target should therefore be considered a lower-priority experimental hypothesis rather than a predicted mechanism.

Predicted ADMET descriptors did not validate the docking ranking. Berberine’s quaternary cationic structure, formulation-dependent exposure, and known pharmacokinetic complexity require experimental assessment. Niclosamide’s limited solubility and exposure likewise make formulation-oriented work important; its repurposing interest arises primarily from prior experimental anti-H. pylori evidence, not from a small difference in mean docking score. Levofloxacin was used only as a clinically interpretable comparator.

From a pharmacological standpoint, the value of the analysis is the joint description of predicted poses, literature-based biological plausibility, and developability questions, not the demonstration of target engagement. Berberine had the most negative descriptive mean score, but direct translation is limited by its quaternary cationic structure and formulation-dependent exposure. Niclosamide remains a logical choice for targeted anti-H. pylori testing because of prior experimental evidence and a top-ranked urease pose, not because binding was established. Levofloxacin served as a clinical comparator rather than a lead candidate.

Initial urease-inhibition tests, broth microdilution or agar-dilution MIC testing against reference and clinical H. pylori strains, checkerboard assays using standard antibiotics, and cytotoxicity testing on gastric epithelial cell models are practical next steps. Molecular dynamics and post-docking free-energy analyses could be considered for experimentally supported complexes. The resulting pharmaceutical message is deliberately restricted: (i) docking scores provide a workflow-specific ranking, not proof of binding; (ii) interaction maps describe the residue-level contacts of predicted poses; and (iii) ADMET descriptors flag properties requiring experimental pharmacokinetic and formulation assessment. None of these outputs demonstrates therapeutic suitability.

This integrated interpretation is particularly suitable for further laboratory work, including urease inhibition assays, MIC determination against H. pylori strains, cytotoxicity screening, and, if results justify it, molecular dynamics simulations.

Integrated prioritization was expressed narratively rather than in a separate redundant table. Within this limited matrix, berberine produced the most consistent score pattern, niclosamide remained the clearest repurposing candidate for urease-focused testing because of prior experimental evidence, and levofloxacin served as a clinically useful comparator. This ordering is a pragmatic experimental-prioritization proposal and not a validated affinity or potency ranking.

Representative three-dimensional poses supplement the numerical scores by showing the orientations selected by the docking algorithm. The urease and DNA gyrase examples display geometrically plausible predicted accommodation, while the two-dimensional diagrams describe the associated modeled contacts. Visual plausibility is not validation and does not establish affinity. The combined representation is useful only as a transparent record of the hypotheses selected for future testing.

The simultaneous integration of three target classes—urease, a virulence/adaptation enzyme; DNA gyrase A, a replication-related target; and shikimate kinase, an enzyme involved in a specific metabolic pathway—is a key aspect of this work. Compared with a single-protein docking experiment, this multitarget approach provides more information and is more consistent with early-stage pharmacological prioritization. The method is especially suitable for natural and repurposed compounds, for which both biological plausibility and ADMET feasibility should be considered before investing in experiments.

Therefore, the comparative value of the manuscript does not depend on the claim that docking predicts clinical response. Using a shared target panel and a uniform ADMET interpretation, it provides an organized ranking of mechanistically plausible compounds that already have antibacterial or anti-H. pylori relevance. Using computational research to reduce the experimental search space before costly microbiological or enzymatic testing, this framework supports early-stage pharmacological screening.

This research is a preliminary, hypothesis-generating prioritization study. A docking score is an empirical score for a modeled complex, not the binding free energy. Thermodynamically, delta G binding = G complex – (G free protein + G free ligand); the present workflow does not determine the free-state, solvation, desolvation, and entropic contributions needed for that quantity. A ligand with a favorable docked-complex score may therefore still prefer solution, self-associate, or fail to bind, and virtual-screening workflows can produce substantial false-positive rates (Ivanov 2026a, 2026b). No formal redocking or enrichment validation, MD simulations, molecular mechanics/generalized Born surface area (MM/GBSA), or molecular mechanics/Poisson–Boltzmann surface area (MM/PBSA) analyses were performed. Receptor flexibility, solvent effects, alternative protonation microstates, and induced fit were not fully modeled. The DNA gyrase A model also does not reproduce the DNA–enzyme–quinolone ternary complex. Consequently, the study demonstrates neither binding, target engagement, enzyme inhibition, bacterial killing, nor therapeutic efficacy. The proposed next steps are urease-inhibition testing, MIC and minimum bactericidal concentration (MBC) determination in reference and clinical H. pylori strains, combination testing, cytotoxicity assays, and, if justified by experimental results, MD and post-docking free-energy analyses.