Integrated In-Silico Prioritization of Anti-Diabetic Candidates from Helianthus Tuberosus Using Bibliometrics, Target Prediction, Docking, MD, and ADMET Profiling

2.1. Data Collection and Processing

This study conducted a comprehensive bibliometric analysis to evaluate the medicinal potential of H. tuberosus by identifying the most frequently used keywords in studies that have assessed H. tuberosus. Data were systematically retrieved from the Web of Science and Scopus databases in February 2025. The search strategy, presented in Fig. (1), involved querying article titles, abstracts, and keywords using a combination of Boolean operators and wildcard symbols to capture relevant literature, as outlined in Table S1. No restrictions were placed on the publication year, ensuring a comprehensive dataset. To maintain consistency, only articles published in English were included in the analysis. The metadata from both databases were exported in BibTeX and plain text formats, respectively. These datasets were then imported into RStudio [version 2024.12.1 Build 563] for integration and analysis. After removing duplicates and normalizing the metadata, a refined dataset of 1,049 unique documents was obtained from the initial pool of publications (Table S1).

2.2. Finding the Antidiabetic Compounds Resulting from Fermentation and Reported in different Parts of H. tuberosus

Anti-diabetic candidates were collected in two categories: [i] fermentation-derived metabolites produced by gut microbial metabolism of inulin [short-chain fatty acids and organic acids], and [ii] phytochemicals reported in H. tuberosus with documented anti-diabetic or glucose-modulating effects [13, 14]. Compounds were compiled from published reviews and experimental studies and prioritized based on bibliometric frequency and reported biological relevance.

2.3. Predicting the Targeted Proteins of Anti-Diabetic Compounds derived from H. tuberosus

We chose the SwissTargetPrediction online tool, a knowledge-based approach to predict new targets of an uncharacterized molecule or secondary targets for a known molecule [15], to predict the best possible protein targets of phytochemicals reported in different parts of H. tuberosus with antidiabetic characteristics. In SwissTargetPrediction, predictions are made based on similarity to known ligands of target proteins, and the scoring system is designed to rank the likelihood of a compound binding to a specific target [16]. While the tool does not always provide a strict cutoff value, we considered scores close to or above 0.5 to indicate reasonable confidence in predictions based on previous surveys [17]. Target prediction was restricted to Homo sapiens. The top-ranked predicted targets were retained, and proteins with a prediction probability ≥0.5 were considered for downstream analysis, consistent with prior computational pharmacology studies.

2.4. Preparing Ligands and Predicted Targeted Proteins for Molecular Docking Analysis

The 3D structures of anti-diabetic compounds derived from both fermentation and other parts of H. tuberosus and the predicted targeted proteins were obtained from PubChem [National Center for Biotechnology Information [NCBI]. [2025]. PubChem Compound Summary for anti-diabetic compounds derived from fermentation and other parts of H. tuberosus], PubChem, and the Protein Data Bank [PDB] [18] online databases, respectively. The UniProt online database [19] was chosen to obtain the 3D structures of proteins whose 3D structures are not mentioned in the PDB.

2.5. Molecular Docking Analysis

Molecular docking analysis was performed between anti-diabetic compounds derived from fermentation and other parts of H. tuberosus and predicted targeted proteins using UCSF Chimera [20] and AutoDock Vina [21]. The highest binding affinity [Kcal/mole] was considered the most appropriate conformation of the interaction between the ligand and protein. The ligands and targeted proteins with the highest binding affinity were selected for molecular dynamics simulations. Protein structures were prepared by removing water molecules and heteroatoms, adding polar hydrogens, and assigning partial charges. Docking grids were centered on the active sites of each protein, and AutoDock Vina was run using default exhaustiveness. The pose with the lowest binding free energy and biologically plausible interactions was selected for MD simulation.

2.6. Molecular Dynamics Simulation

Molecular dynamics simulations for the ligand-protein conformations with the highest binding affinity were performed using GROMACS [20] in the Linux environment with the help of UCSF Chimera. The duration of the molecular dynamics simulation was 10 nanoseconds [ns], and it was performed in a molecular dynamics run of 5,000,000 steps. Conformations with an RMSD value of less than 0.2 nm are generally considered acceptable for a high-quality model for molecular simulations and docking applications. RMSD values were calculated for ligand heavy atoms relative to the initial docked conformation.

2.7. The Bioavailability Assay

The bioavailability of anti-diabetic compounds derived from both fermentation and other parts of H. tuberosus, with the most stable ligand-protein interactions with the predicted targeted protein, was assessed using the SwissADME online tool [23]. Based on the principles of the SwissADME online tool, various criteria were evaluated for assessing the bioavailability of components derived from H. tuberosus [23]. Ultimately, compounds with a bioavailability score of 0.5 or greater were considered suitable components for the drug prediction assay.

2.8. Drug Prediction Assay

The anti-diabetic compounds derived from both fermentation and other parts of H. tuberosus, with the most stable ligand-protein connection and the predicted targeted protein, and a proper bioavailability score, underwent evaluation using two online tools: ProTox 3 [24] and Way2Drug [25]. ProTox3 demonstrated the detailed molecular features, LD50, toxicity class, and cellular pathways that can be affected by the toxicity of this compound, which could be utilized as a future anti-diabetic drug. Toxic doses are often given as LD50 values in mg/kg body weight. The LD50 is the median lethal dose, meaning the dose at which 50% of test subjects die upon exposure to a compound. Toxicity classes are defined according to the Globally Harmonized System of Classification and Labeling of Chemicals [GHS]. LD50 values are given in [mg/kg] [24]:

Class I: fatal if swallowed [LD50 ≤ 5]

Class II: fatal if swallowed [5 < LD50 ≤ 50]

Class III: toxic if swallowed [50 < LD50 ≤ 300]

Class IV: harmful if swallowed [300 < LD50 ≤ 2000]

Class V: may be harmful if swallowed [2000 < LD50 ≤ 5000]

Class VI: non-toxic [LD50 > 5000]

Moreover, Way2Drug demonstrated which molecular or cellular pathways will be affected in case of toxicity with the mentioned compounds as an anti-diabetic drug [25].

2.9. Data Visualization

Discovery Studio Visualizer software was used to present molecular docking results in 3D. The results of the RMSD condition of ligands were demonstrated using Grace software [Grace Software, version 5.1.22, Batwing, B. [Developer], 2021. Available at: http://plasma-gate. weizmann.ac.il/Grace/]. Moreover, Cytoscape software [26] was used to depict the molecular pathways by which compounds derived from H. tuberosus exert their anti-diabetic roles.

3.1. The Results of Keyword Analysis from the Bibliometric Analysis

Figure 2 presents the frequency of key terms used in scientific publications related to the medicinal potential of H. tuberosus. The most commonly mentioned terms include: “Jerusalem artichoke” (271 occurrences), “Helianthus tuberosus” (220), “article” (191), “Helianthus” (142), “fermentation” (133), “nonhuman” (129), “inulin” (118), “sunflower” (114), “metabolism” (102), and “controlled study” (88). Among the keywords mentioned, Jerusalem artichoke (Helianthus tuberosus), Helianthus, nonhuman, sunflower, and controlled study are general keywords that do not focus on a specific scientific field. However, the three keywords of fermentation, inulin, and metabolism highlight the focus of scientists on evaluating various medical applications of H. tuberosus.

3.2. Compounds Resulting from Fermentation and Reported in different Parts of H. tuberosus

Acetate, Butyrate, Ethyl acetate, Lactate, Propionate, and Succinate were compounds derived from the fermentation of inulin in the gastrointestinal [GI] tract. Besides, Faradiol, 5-Feruloylquinic acid, Ferulic acid, Sulferein glycoside, and Vanillin are phytochemicals in the structure of H. tuberosus with anti-diabetic features. Tables 1 and S2 demonstrate detailed information about these molecules.

3.3. Prediction of the Best Possible Targeted Proteins that Phytochemicals and Compounds Reported in different Parts of H. tuberosus, with Antidiabetic Characteristics, can Target

Figure S1 displays different proteins that can be targeted by phytochemicals and compounds reported in other parts of H. tuberosus and the 2D structures of the mentioned compounds derived from this plant. Based on the information in Figure S1, butyrate, ferulic acid, faradiol, and succinate can target proteins in humans with an acceptable probability.

Additionally, Table 2 highlights the proteins that can be targeted by phytochemicals and compounds reported in various parts of H. tuberosus with antidiabetic properties in the human body. Moreover, Table 2 illustrates the diverse roles of these proteins and the molecular pathways through which they can impact diabetes. PTP1B is a negative regulator of insulin receptor signaling, and its inhibition improves insulin sensitivity. HDAC3 is implicated in metabolic inflammation and endothelial dysfunction in diabetes. Carbonic anhydrase isoforms, including CA XII, are involved in metabolic and vascular regulation in diabetic conditions. EGLN1 [PHD2] regulates hypoxia-inducible signaling, which is relevant to angiogenesis and wound healing in diabetes.

3.4. Molecular Docking Analysis

Molecular docking analysis demonstrated that butyrate, ferulic acid, faradiol, and succinate could target proteins in the human body in an in-silico environment (Fig. 3). Moreover, Figure 3 displays detailed information on the molecular interactions among butyrate, ferulic acid, faradiol, and succinate, as well as the potential target proteins in the human body.

3.5. The Results of the Molecular Dynamics Simulation

Based on our findings from the molecular dynamics simulation, Butyrate and Ferulic acid, with average RMSD values of less than 0.5 nm, displayed stable interactions with HDAC3 and Carbonic anhydrase XII, respectively. Thus, these two compounds underwent a bioavailability assay (Table 3).

3.6. The Results of the Bioavailability Assay

Based on data from the SwissADME online tool, the bioavailability scores for Butyrate and Ferulic acid are acceptable at 0.85. In addition, Figure 4 displays the Bioavailability Radar for Butyrate and Ferulic Acid, which highlights the key molecular features of these two compounds as therapeutic agents in the human body. Contrary to Butyrate, different features of Ferulic acid are not entirely located in the pink area.

3.7. Results of the ProTox3 and Way2Drug Analysis

Based on the analysis of ProTox3, butyrate is categorized as Toxicity Class 3 as an anti-diabetic drug, with an LD50 of 91 mg/kg (Fig. 5). Figure 5 also depicts ProTox3's prediction accuracy. Besides, the analysis using the ProTox3 online tool has also shown that the toxicity of butyrate as an oral anti-diabetic drug can adversely affect the blood-brain barrier [BBB] (Fig. 5). Moreover, in detail, the Way2Drug online tool demonstrated the different side effects of butyrate as an oral anti-diabetic drug (Table S3).

3.8. Various Steps of the Present Drug Design and Drug Discovery

Based on the steps outlined in the present review, Table 4 lists the consecutive steps involved in drug design and discovery.

4.1. Interpretation of Butyrate Toxicity Predictions

Although ProTox3 classified butyrate as Toxicity Class 3 at high predicted doses, butyrate is a naturally occurring short-chain fatty acid produced endogenously in the colon through microbial fermentation of dietary fiber. The predicted LD50 reflects the acute toxicity of an isolated, systemically available compound rather than physiological colonic exposure. In vivo, butyrate is largely metabolized locally by colonocytes, and systemic concentrations depend on dose, formulation, and route of administration. Therefore, the observed in silico toxicity highlights the importance of dose optimization and formulation strategies rather than excluding butyrate as a candidate.

4.2. Strengths and Limitations

This study has several notable strengths. First, it employed an end-to-end, data-driven pipeline that integrated bibliometric analysis, target prediction, molecular docking, short-timescale molecular dynamics [MD] simulations, pharmacokinetic and drug-likeness evaluation, and in silico toxicity profiling. This comprehensive workflow reduces reliance on any single method and improves the reliability of candidate selection. Second, by focusing on natural products from Helianthus tuberosus, the study highlights a widely available plant source with both nutraceutical and pharmaceutical potential, thereby enhancing translational feasibility. Third, the use of publicly available databases such as Web of Science, Scopus, PubChem, PDB, and UniProt, combined with open-access computational tools like SwissTargetPrediction, AutoDock Vina, GROMACS, SwissADME, ProTox3, and Way2Drug, strengthens methodological transparency and reproducibility. Another strength lies in the triangulation of results across docking scores, MD stability [RMSD], and ADME-toxicity filters, which converged on a small number of tractable lead compounds, namely butyrate and ferulic acid, thus reinforcing confidence in their therapeutic promise. Furthermore, by identifying multiple target proteins and pathways, including HDAC3, CA isoforms, and PTP1B, the study adopts a systems-level view that supports hypothesis generation for multi-target interventions in diabetes. Finally, the extensive use of visual outputs, such as molecular interaction diagrams, bioavailability radars, and pathway schematics, enhances the interpretability of results and facilitates their application in subsequent experimental design.

Despite its strengths, the study has several limitations. The initial compound selection was guided by bibliometric trends, which introduces a degree of selection bias and may exclude less-studied but potentially valuable molecules. The analysis was entirely in silico, meaning that the findings still require experimental confirmation in biochemical assays, cellular systems, and in vivo diabetic models to validate efficacy and safety. Computational approaches also have inherent constraints; for instance, docking was performed with a single engine, and MD simulations were limited to 10 nanoseconds, which restricts conformational sampling. Protein structures were derived from available PDB entries or modeled databases, which may not fully represent physiological states, cofactors, or active-site conditions. In terms of chemistry, the screened library included only a limited set of fermentation products and phytochemicals from H. tuberosus, without exhaustive exploration of stereoisomers, metabolites, or microbiome-derived derivatives. Predictive tools such as SwissADME, ProTox3, and Way2Drug also have limited applicability domains, and their results may not fully capture organ-specific toxicity, off-target effects, or clinical pharmacokinetics. Importantly, the broader biological context was not modeled, including membrane permeability, transporter activity, protein-protein interactions, and hormonal regulation of glucose metabolism. Moreover, the biological roles of HDAC3 and carbonic anhydrases are complex and context-dependent, meaning that therapeutic outcomes could vary across tissues and disease states. Finally, while butyrate emerged as a promising candidate, its predicted moderate toxicity, volatility, and potential blood-brain barrier liability highlight translational challenges that must be addressed through formulation strategies and careful dose optimization.