Integrative Network Pharmacology and Molecular Docking Approaches in Herbal Medicine Research. A Systematic Review of Applications, Advances, and Translational Potential

2.1 Study design

A systematic review was conducted in accordance with PRISMA 2020 and the PRISMA extension for Scoping Reviews (PRISMA-ScR). The protocol specified the research question, eligibility criteria, information sources, search strategy, screening processes, data extraction fields, quality appraisal approach, and synthesis plan before data collection commenced.

2.2 Eligibility criteria

Studies were included if they met specific methodological and time-based criteria. Eligible papers were original, peer-reviewed research articles published between January 1, 2010 to 1^st August, 2025. Each study was required to use both network pharmacology and molecular docking as part of its research approach, and the interventions had to involve herbal medicines, traditional plant formulations, or isolated phytochemicals. Studies also needed to report measurable scientific outputs such as compound–target interactions, protein network results, pathway or enrichment analysis, or docking scores showing ligand–receptor binding. For consistency and accurate interpretation, only studies published in English were considered. Studies were excluded if they did not use both network pharmacology and molecular docking together. Review articles, systematic reviews, meta-analyses, commentaries, letters, conference abstracts, and unpublished work were not included. Research focusing only on synthetic drugs without any herbal component was excluded. In addition, non-English publications and grey literature, such as theses and institutional reports, were not considered in this review.

2.3 Information sources and search strategy

A comprehensive literature search was conducted on 1^st August 2025 to identify eligible studies. Three major scientific databases were searched including PubMed, Scopus, and Web of Science. A total of 83 records were retrieved across the three databases (PubMed = 26, Scopus = 10, Web of Science = 47). All articles were retrieved as CSV file and imported into Rayyan for systematic screening, duplicate detection, and management. During the de-duplication process, 9 duplicate records were detected. 4 duplicates were resolved and 5 was removed, leaving 78 articles for title and abstract screening. The search strategy combined controlled vocabulary terms and free-text keywords related to network pharmacology, molecular docking, and herbal medicine. The PubMed search string was: (“Network Pharmacology” [Title/Abstract] OR “Systems Pharmacology” [Title/Abstract]) AND (“Molecular Docking” [Title/Abstract] OR “In Silico Docking” [Title/Abstract]) AND (“Herbal Medicine” [Title/Abstract] OR “Medicinal Plants” [Title/Abstract] OR “Phytochemicals” [Title/Abstract]) AND (“2010/01/01” [Date-Publication]: “2025/12/31” [Date-Publication]). Equivalent search terms were adapted for Scopus and Web of Science to ensure consistency while accommodating database-specific syntax.

2.4 Study selection and screening

The study selection process followed a structured and transparent approach consistent with PRISMA-ScR guidelines. After the initial search and de-duplication, 78 articles were screened for title and abstract. Two reviewers independently assessed each article against the defined eligibility criteria. Articles that met the inclusion criteria were selected for full-text screen. During the screening process, any disagreements between the two reviewers were addressed through discussion. When consensus could not be reached, a third reviewer was consulted to resolve the discrepancy. Full-text articles were then examined to confirm compliance with the predefined inclusion criteria. Decisions at each stage of the screening process, including reasons for exclusion, were documented systematically. Out of the screened records, 36 studies met all criteria and were included in the final synthesis as depicted by Figure 1.

Figure 1. PRISMA flow chart.

2.5 Data extraction

Data extraction was carried out using a structured and predefined data collection framework drafted in excel (version 2019) to ensure consistency and completeness. Key study characteristics and methodological details were systematically extracted from each included article. Extracted information included bibliographic data (author, year, journal, and country), herbal species or formulations investigated, plant part used, extraction or preparation method (where applicable), and the class of bioactive compounds assessed. Technical information relevant to methodologies and computational workflows was also collected. This included the databases used for compound identification and target prediction, software and algorithms applied in network construction, molecular docking tools and scoring functions, validation strategies, and major biological pathways or molecular targets identified. In addition, details related to experimental design, such as in vitro or in vivo validation, were captured when reported. The extraction process was conducted independently by two reviewers, and extracted data were cross-checked for accuracy. Any discrepancies were resolved through joint review and consensus discussion. Where information was incomplete or ambiguous, efforts were made to confirm details directly from study text and supplementary materials to ensure accurate interpretation.

2.6 Quality assessment

Quality assessment focused on methodological transparency, rigor of computational approaches, and validity of outcomes. Studies were evaluated for reporting of databases, compound selection, target identification, network construction, docking protocols, scoring methods, and validation. Adherence to accepted standards in network pharmacology and molecular docking, including justification of tools, was assessed. Consistency with biological evidence and clarity of key outputs (compound–target interactions, pathways, docking scores) were also considered. Two reviewers independently appraised studies, with disagreements resolved by consensus or a third reviewer. Only methodologically sound studies were included.

2.7 Data synthesis and analysis

A descriptive synthesis summarized study characteristics, herbal taxa, phytochemicals, disease indications, targets, and pathways. Frequencies and percentages were calculated for databases, prediction tools, network software, docking platforms, validation methods, and pathway resources, with totals normalized to 100% where applicable. Temporal trends (2010–2025), country distribution, and therapeutic areas were also summarized. Network maps of herb–compound–target–pathway relationships were generated where data permitted.

3.1 Study characteristics

A total of 36 studies met the inclusion criteria and were incorporated into the final synthesis. The studies were published between 2013 and 2025, with a significant increase in publications from 2020 onward. Majority of the studies were conducted in Asian countries, particularly China (n = 20, 55.6%)^4,10,11 and India (n = 9,¹² 25.0%), followed by Pakistan, Saudi Arabia, Egypt,^13,14 Ethiopia,⁸ Indonesia, and Taiwan.¹⁵ Significant number of studies employed fully in silico approaches integrating network pharmacology and molecular docking (n = 22, 61.1%).¹² The remaining studies combined computational pipelines with experimental laboratory validation, including in vitro assays (n = 8, 22.2%) and in vivo animal models (n = 6, 16.7%). Sample sizes for experimental arms varied, with most in vivo studies utilizing murine models, while in vitro analyses frequently employed human cancer cell lines and macrophage systems as summarized in Table 1. Plant materials analyzed varied widely and included whole plants, roots, rhizomes, barks, fruits,^10,12 seeds, leaves, and formulated polyherbal preparations. Solvents used for extraction, where specified, included methanol,¹⁶ ethanol, chloroform, and n-hexane; however, solvent reporting was inconsistent across studies, with nearly half not specifying extraction media. Both single-herb studies (e.g., Centella asiatica, Piper longum, ¹⁰ Saussurea lappa) and polyherbal formulations (e.g., Huangqin Tang, Dashamoola, HSXFF, Krishnadi Churna, Qi-Enriching and Blood-Tonifying herbs) were reported. Biological validation where present included evaluation in disease-specific cell lines such as MCF-7 and A549, immune cells such as RAW264.7 macrophages and human peripheral blood cells, and rodent models of neurodegeneration, inflammation, ulcerative colitis, and metabolic dysfunction. Most purely computational studies validated predicted targets and pathways using publicly available protein structures, genomic databases, and docking or molecular dynamics outputs.

3.2 Temporal distribution of included studies

The temporal distribution of studies indicates a steady increase in the use of network pharmacology and molecular docking in herbal medicine research presented in Figure 2. Early activity was limited, with only 2.78% of studies published in 2013¹⁵ and 5.56% in 2016, reflecting the initial stage of adoption. A small increased appeared in 2020 (2.78%), followed by more noticeable growth in 2021 (8.33%), marking the beginning of wider implementation of computational methods in the field. A significant expansion occurred from 2022 onward, where publication frequencies reached 19.44% in 2022 and 13.89% in 2023.¹¹ The highest proportion of studies was observed in 2024 (30.56%),^19,11 and a substantial number continued to be published in 2025 (16.67%).⁴

Figure 2. Temporal distribution of studies applying network pharmacology and molecular docking in herbal medicine research.

The temporal distribution shows studies were reported in 2013 (2.78%), 2016 (5.56%), 2020 (2.78%), 2021 (8.33%), 2022 (19.44%), 2023 (13.89%), 2024 (30.56%), and 2025 (16.67%).

3.3 Distribution of database categories reported

The included studies utilized different computational databases to identify phytochemicals, predict molecular targets, and characterize disease-associated genes. These databases included herbal/phytochemical repositories, chemical-structure databases, target-prediction platforms, and disease–gene resources. Additional tools were used for ADME–toxicity screening and protein interaction analysis.

3.3.1 Herbal/phytochemical repositories

Across the included studies, TCMSP was the most frequently used herbal database, accounting for 52.9% of all phytochemical database citations as presented in Figure 3. This was followed by IMPPAT (17.6%) and TCM Database@Taiwan (11.8%),¹⁵ whereas HERB, ETCM, and NPASS were each used in 5.9% of studies. These findings suggest that most researchers rely on widely established Chinese herbal databases, with relatively lower utilization of newer or region-specific phytochemical repositories.

Figure 3. Distribution of herbal and phytochemical databases used for compound retrieval in included studies.

The figure illustrates the frequency of herbal database utilization across the reviewed literature (n = 17 mentions). TCMSP was the most commonly employed platform (52.9%), followed by IMPPAT (17.6%) and TCM Database@Taiwan (11.8%). HERB, ETCM, and NPASS were each used in 5.9% of studies.

3.3.2 Target-prediction platforms used in included studies

SwissTargetPrediction was reported in 36.36% of studies (n = 8), representing the most frequently used platform for target prediction ( Table 2). DIGEP-Pred, SEA (Similarity Ensemble Approach), and non-specified or model-based approaches were each reported in 9.09% of studies (n = 2 each), indicating moderate use across the included studies. PharmMapper, Way2Drug, SuperPred, ImaGEO, STITCH (used for target assignment), STRING (used for protein interaction-based target assignment), TTD, and PharmGKB were each reported in 4.55% of studies (n = 1 each).

The results show that SwissTargetPrediction was the most commonly employed platform (36.36%), followed by DIGEP-Pred, SEA, and non-specified computational approaches (each 9.09%). PharmMapper, Way2Drug, SuperPred, ImaGEO, STITCH, STRING, TTD, and PharmGKB were each used in 4.55% of studies, indicating selective adoption of specialized tools for structure-based prediction, transcriptomic inference, and protein-interaction mapping.

3.3.3 Chemical structure databases

Chemical structure databases were used to support compound identification and structural verification across the included studies as depicted in Figure 4. PubChem was the most frequently used resource (31.25%), followed by SwissADME (18.75%). Other tools including ChEMBL,²² admetSAR, ChemDraw,²³ NIST, Dictionary of Natural Products, and CDRUG were each reported in 6.25% of cases. This distribution reflects a primary reliance on established public chemical repositories such as PubChem, with supplementary use of cheminformatics and property-prediction tools for compound validation.

Figure 4. Chemical structure databases used for compound characterization across included studies.

The result show that pubChem was most commonly employed (31.25%), followed by SwissADME (18.75%), with ChEMBL, admetSAR, ChemDraw, NIST, Dictionary of Natural Products, and CDRUG each used in 6.25% of studies.

3.3.4 Disease-gene annotation databases used across included studies

Disease-gene retrieval in the included studies that primary relied on GeneCards (44.44%), followed by OMIM and DisGeNET (22.22% each) as depicted by Figure 5. GEO, NCBI, and HPO were each utilized in 5.56% of studies. This pattern indicates a primary dependence on comprehensive disease-gene compendiums, particularly GeneCards,¹⁶ supplemented by clinically annotated resources (OMIM) and disease-association platforms (DisGeNET). Use of curated transcriptomic repositories (GEO)²¹ was less common, reflecting fewer studies integrating omics-driven disease gene mining.

Figure 5. Disease-gene annotation databases used across included studies.

The figure results of the Disease-gene retrieval revealed GeneCards as the most frequently used disease-gene source (44.44%), followed by OMIM and DisGeNET (22.22% each). GEO, NCBI, and HPO accounted for 5.56% each, indicating selective incorporation of transcriptomic and phenotype-ontology sources.

3.3.5 Protein annotation and interaction databases used in the included studies

Across the included studies, protein annotation and interaction resources were used to support target validation and functional characterization as presented in Figure 6. UniProt/UniProtKB was the most frequently applied database (62.5%), indicating its central role in annotating protein targets and linking phytochemicals to biological functions. BindingDB appeared in 25.0% of studies, primarily for retrieving experimental protein ligand interaction data, while STITCH was used in 12.5% of studies for identifying chemical protein interaction networks. Together, these tools provided essential support for confirming target relevance and mapping molecular interaction profiles in network-pharmacology workflows.

Figure 6. Protein annotation and interaction databases used in the included studies.

The results show that UniProt/UniProtKB was the most widely used platform, accounting for 62.5% of studies, followed by BindingDB (25.0%) and STITCH (12.5%). These resources were primarily applied for protein functional annotation, target confirmation, and mapping chemical–protein interaction networks within network-pharmacology pipelines.

3.3.6 Computational multi-database integrators

Among the included studies, some applied multi-database integration approaches that combine chemical, genomic, proteomic, and pathway data within a single workflow. These approaches were used to improve compound identification, increase confidence in target prediction, and support network construction. Reported platforms included HERB-linked compound and serum-metabolite integration frameworks, the Traditional Chinese Medicine Integrated Platform (TCMIP v2.0), and multi-omics pipelines integrating drug–target, protein–protein interaction, and pathway data (e.g., HERB, TCMIP v2.0), each reported once (12.5%). Other studies used combined database approaches for specific analyses. These included pathway enrichment (e.g., Metascape), compound validation using multiple chemical databases (e.g., PubChem, ChEMBL, Dictionary of Natural Products), and integration of experimental data (e.g., LC-MS/GC-MS) with disease–gene databases (e.g., DrugBank, OMIM, DisGeNET, GeneCards). One study also used a combination of a herbal database and viral-target literature to support antiviral drug discovery.

3.3.7 Literature-derived compound sources

Reported sources included literature-derived compound datasets integrated with filtering tools (e.g., PreADMET, Osiris Property Explorer) and manual literature searches combined with phytochemical and ADME databases (e.g., IMPPAT, SwissADME). Some studies cross-validated literature-reported compounds using target and disease databases (e.g., SwissTargetPrediction, GeneCards, DisGeNET). In several cases, studies developed in-house compound libraries based entirely on literature mining, while others used literature-supported compound lists alongside established herbal databases (e.g., TCM Database@Taiwan). Additional approaches included literature-based repurposing datasets for antiviral compounds (e.g., SARS, MERS, SARS-CoV-2) and disease-specific compound extraction from published studies (e.g., prostate cancer data). Overall, ten studies (12.5%) incorporated literature-derived phytochemical sources as part of their compound identification strategy.

3.4 Therapeutic domains in the included studies

Inflammatory and immune-mediated conditions accounted for 14.04% of studies, followed by oncology-related investigations. Cancer-focused studies included breast cancer (8.77%), lung cancer (5.26%), colon/colorectal cancer (3.51%), prostate cancer (3.51%), hepatocellular carcinoma (1.75%), cervical cancer (1.75%), and unspecified cancer types (10.53%). COVID-19-related studies accounted for 7.02% of the dataset. Neurological and psychiatric conditions included Alzheimer’s disease (1.75%), Parkinson’s disease (1.75%), and anxiety/depression (3.51%). Metabolic and cardiovascular conditions comprised diabetes (1.75%), hypertension (1.75%), vascular calcification (1.75%), and oxidative stress-related disorders (1.75%). Gastrointestinal immune disorders, including ulcerative colitis (3.51%) and Crohn’s disease (1.75%), were also reported ( Table 3).

The results show that inflammatory and immune-mediated disorders were the most frequent focus (14.04%), followed by cancer-related investigations (breast cancer 8.77%, cancer-unspecified 10.53%, lung cancer 5.26%, colon and prostate cancer each 3.51%, and hepatocellular and cervical cancer each 1.75%). COVID-19 accounted for 7.02% of studies, while neurological disorders (Parkinson’s and Alzheimer’s disease), metabolic and cardiovascular conditions (diabetes, hypertension, vascular calcification), and gastrointestinal immune diseases (ulcerative colitis, Crohn’s disease) each represented 1.75–3.51% of the literature.

3.5 Network-pharmacology software used in included studies

Across the included studies, Cytoscape emerged as the primary network-pharmacology platform, appearing in 63.89% of cases as presented in Table 4. Among these, versions v3.7.2 and v3.8.2²³ were most frequently reported (8.33% each), while other versions including v3.7.0,²⁴ v3.10, v3.9.0, v3.9.1, v3.7.1, and v3.2 were each identified in 2.78% of studies. Nevertheless, version information was omitted in 30.56% of Cytoscape-using studies, indicating incomplete reporting of software details. Beyond Cytoscape, STRING²³ was the next most commonly used platform (22.22%), followed by AutoDock-based pipelines (11.11%) and STITCH (5.56%). Less frequently adopted tools included GEPIA2, KEGG/GO^16,20 enrichment platforms, Python NetworkX, and machine-learning-driven approaches (each 2.78%). Additionally, 19.44% of studies applied network-pharmacology workflows without specifying the software used, highlighting variability and reporting gaps in computational methodology.

The results indicate different computational workflows, with multiple versions of Cytoscape widely used (8.33% each for v3.7.2 and v3.8.2; 30.56% unspecified). STRING was applied in 22.22% of studies, and AutoDock-based pipelines in 11.11%. STITCH and other tools including GEPIA2, KEGG/GO enrichment tools, Python NetworkX,⁸ and machine-learning models each appeared in 2.78% of studies. Additionally, 19.44% of studies referenced network-pharmacology pipelines without specifying software, highlighting reporting variability.

3.6 Distribution of pathway and enrichment analysis tools among included studies

The distribution of pathway and enrichment tools shows that KEGG Pathway Analysis accounted for 37.88%, while Gene Ontology (GO) accounted for 28.79% and DAVID for 16.67%. Reactome and BioCarta were each reported at 3.03%. ShinyGO, KOBAS, BBID, WikiPathways, GOBP (GO Biological Process), Comparative Toxicogenomics Database network enrichment, and enrichment factor (EF) analysis were each reported at 1.52%, indicating that these tools were applied individually across different studies ( Table 5).

Distribution of pathway and enrichment analysis tools used across included studies. The results show that KEGG (37.88%) and Gene Ontology (28.79%) were the most frequently applied enrichment resources, followed by DAVID (16.67%). Other tools, including Reactome, BIOCARTA, ShinyGO, KOBAS, and WikiPathways, accounted for smaller shares (≤3.03%), demonstrating limited adoption beyond core KEGG- and GO-based platforms.

3.7 Distribution of core protein targets identified across included studies

The analysis reveals a concentration of research efforts on key molecular regulators implicated in cancer progression, inflammation, cell survival, apoptosis, and angiogenesis as presented in Table 6. The most frequently investigated targets included ESR1 (27.78%), EGFR (25.00%),²⁰ and AKT1 (22.22%), reflecting strong emphasis on hormone-dependent and growth-factor-mediated signaling pathways.²³ Inflammation-related proteins such as TNF (19.44%), IL6 (11.11%), and IL1B (8.33%),¹⁷ alongside apoptosis-linked mediators (CASP3 and PTGS2, each 16.67%) were also prominently represented. Similarly, tumor suppressors and stress-response proteins (TP53 and HSP90AA1, each 13.89%) and angiogenesis and matrix-remodeling factors (VEGFA and MMP9, each 13.89%) appeared consistently across studies.

Overall, these patterns highlight a research emphasis on interconnected oncogenic and immuno-inflammatory pathways, demonstrating that herbal-network-pharmacology investigations predominantly target biologically relevant hubs associated with cancer, metabolic disorders, and inflammation-related diseases.

The results show that ESR1 (27.78%), EGFR (25%), and AKT1 (22.22%) were the most frequently investigated proteins, followed by inflammation-related factors such as TNF (19.44%) and CASP3/PTGS2 (16.67% each). Targets predominantly reflected pathways of cancer progression, inflammatory signaling, and metabolic regulation.

3.8 Distribution of docked phytochemicals across included studies

Across the included studies, phytochemical docking was more prevalence with focused on flavonoid compounds, with quercetin (25%),²⁴ kaempferol (13.9%),¹⁶ apigenin (11.1%), and luteolin (8.3%) being the most frequently evaluated ligands as presented in Table 7. Phytosterols such as β-sitosterol (8.3%) and polyphenols including gallic acid (5.6%) and EGCG (5.6%) also appeared consistently. The remaining compounds (75%) were evaluated in single studies, indicating ongoing screening of diverse herbal metabolites spanning alkaloids, terpenoids, lignans, catechins, and phenolic acids.

Across the included studies, flavonoids dominated docking analyses, with quercetin (25%), kaempferol (13.9%), apigenin (11.1%), and luteolin (8.3%) being the most frequently evaluated compounds. Other commonly utilized phytochemicals included β-sitosterol (8.3%), gallic acid (5.6%), and EGCG (5.6%). Alkaloids, phenolic acids, steroidal compounds, and saponins appeared less frequently, typically in disease-specific screening contexts.

3.9 Distribution of reported binding affinities across included studies

The distribution of reported binding affinities across the included studies shows that values ≤ −9.0 kcal/mol accounted for 19.4%, while values between −8.0 and −8.9 accounted for 16.7%. Affinity ranges of −7.0 to −7.9, −6.0 to −6.9, and −5.0 to −5.9 each accounted for 13.9%. Binding affinities greater than −5.0 accounted for 5.6%. Studies reporting mixed affinity ranges across −3 to −11.6 accounted for 8.3%, while unclear or unreported values also accounted for 8.3% ( Table 8).

Across the included studies (n = 36), the majority of phytochemicals demonstrated strong binding affinity, with 19.4% reporting very strong docking scores (≤ −9.0 kcal/mol) and 16.7% showing strong affinities (−8.0 to −8.9 kcal/mol). Moderate binding was observed in ~28% of studies, while weak/variable affinity (> −5 kcal/mol) was uncommon. Approximately 8.3% of studies did not specify binding values.

3.10 Distribution of key molecular pathways identified across included studies

The analysis showed that PI3K–Akt signaling accounted (16.90%), followed by MAPK/ERK (14.08%)²⁴ and NF-κB signaling (12.68%).²¹ Immune and cytokine-related pathways, including TNF, IL-17, and TLR cascades, represented 11.27% of reported pathways. Other reported mechanisms included Wnt/β-catenin signaling (7.75%), apoptosis pathways (7.04%), p53-mediated DNA-damage response (6.34%), and VEGF-linked angiogenesis pathways (6.34%). Pathways associated with oxidative stress (AGE–RAGE; 5.63%) and hypoxia response (HIF-1; 4.93%) were also commonly evaluated. In contrast, neurotransmission-related pathways (4.23%) and antiviral signaling mechanisms (2.82%) were less frequently reported ( Figure 7).

Figure 7. Distribution of key molecular pathways identified across included studies.

Distribution of molecular signaling pathways investigated across included studies based on total pathway mentions (N = 142). PI3K–Akt (16.90%), MAPK/ERK (14.08%), and NF-κB (12.68%) pathways accounted for the largest proportions, followed by inflammatory cytokine signaling (11.27%), Wnt/β-catenin (7.75%), apoptosis regulation (7.04%), and p53 and VEGF-related pathways (6.34% each). Oxidative stress (5.63%) and hypoxia responses (4.93%) were moderately represented, while neurotransmission (4.23%) and antiviral signaling (2.82%) were least frequently reported.

3.11 ADMET/drug-likeness tools identified in included studies

Across the included studies, ADMET and drug-likeness screening was commonly integrated into computational herbal research workflows as summarized in Table 9. SwissADME¹⁹ and TCMSP (OB & DL) were the most frequently used tools, each reported in 17.14% of studies. Lipinski’s rule of five was applied in 14.29% of the studies, highlighting consistent screening for oral bioavailability and small-molecule drug-likeness. Other predictive platforms such as admetSAR (8.57%), QikProp (5.71%), ProTox-II (2.86%), and pkCSM (2.86%) were used less frequently, suggesting selective adoption of advanced pharmacokinetic and toxicity profiling tools. DFT-based stability screening and MMGBSA/MDS simulations appeared in isolated studies (2.86% each), indicating emerging integration of quantum and dynamic computational methods. Notably, 28.57% of studies did not report performing ADMET evaluation, underscoring variability in reporting rigor across the literature.

Distribution of ADMET and drug-likeness evaluation tools across included studies. SwissADME and TCMSP screening (each 17.14%) were the most commonly used approaches, followed by Lipinski’s criteria (14.29%) and admetSAR (8.57%). Advanced computational pharmacokinetic tools such as QikProp, pkCSM, and ProTox-II were used less frequently (2.86%–5.71%). A considerable proportion of studies (28.57%) did not specify an ADMET assessment tool.

3.12 Molecular dynamics simulation usage

Across the 36 included studies, 17 studies (47.22%) incorporated molecular dynamics (MD) simulations as part of their computational workflow, while 19 studies (52.78%) did not perform MD simulations as depicted by Figure 8. This indicates that although MD simulation has emerged as a valuable technique for validating ligand–protein interactions and assessing structural stability, its adoption remains moderate, with just under half of studies integrating MD into their network-pharmacology or molecular docking analyses. Studies that employed MD simulations typically ran simulations ranging from 10 ns to 300 ns, often accompanied by complementary analyses such as MM-GBSA binding free energy calculation, RMSD, RMSF, PCA, FEL, and DCCM, demonstrating increasing methodological rigor among advanced studies.

Figure 8. Distribution of molecular dynamics simulation usage among included studies.

The results show that 47.22% (n = 17) of the studies conducted molecular dynamics simulations, while 52.78% (n = 19) did not incorporate MD procedures. This reflects a moderate adoption of MD simulation in natural-product-based computational drug-discovery studies.

3.13 Mechanistic evidence across included studies

Across the included studies, phytochemicals and herbal formulations demonstrated broad pharmacological activity mediated through multiple molecular pathways as summarized in Table 10. The most frequently reported mechanisms involved suppression of inflammatory mediators and immune signaling, particularly inhibition of NLRP3 inflammasome activation, NF-κB signalling, and pro-inflammatory cytokines such as TNF-α, IL-1β, and IL-6. A substantial proportion of studies also described anticancer effects driven by modulation of key oncogenic pathways including PI3K/Akt, MAPK,^10,19 HIF-1α,²⁰ and EGFR, alongside induction of apoptosis, cell-cycle arrest, and caspase activation. Neuroprotective effects were reported through attenuation of oxidative stress, stabilization of neuronal signaling, and inhibition of acetylcholinesterase activity. Additional mechanisms included metabolic and vascular regulation, antiviral activity (particularly SARS-CoV-2 and HIV-related targets), enhancement of intestinal barrier function, and modulation of endocrine signaling pathways. Overall, the results highlight a consistent trend toward multi-target and systems-level regulatory mechanisms, supporting the therapeutic relevance of network-guided phytochemical research.

Natural compounds and herbal formulations demonstrated multi-target therapeutic actions across inflammation, cancer, metabolic regulation, neuroprotection, antiviral responses, gut barrier function, and hormonal pathways. Core mechanisms included suppression of pro-inflammatory mediators, modulation of PI3K/Akt and MAPK pathways, induction of apoptosis, inhibition of oxidative stress, and regulation of immune and neurotransmitter signaling. These findings support the systems-based therapeutic potential of phytochemicals in diverse chronic disease models.

3.14 Validation modality distribution

Across the 36 studies, in silico validation was the most frequent approach (58.33%), followed by combined in vitro + in silico designs (25.00%)¹⁶ as depicted in Figure 9. Fewer studies implemented laboratory animal validation,²⁴ either with in vitro only (2.78%), in vivo + in silico (5.56%),²³ or in vitro + in vivo without computational components (8.33%).^10,19 No study used in vivo validation alone.

Figure 9. Validation modality distribution.

Ethics approval and consent to participate

Not applicable. This study is a systematic review based entirely on previously published research and did not involve human participants, animals, or the collection of new biological samples.