Web of Science Core Collection (WoSCC) is a comprehensive academic database covering over 190 subject areas globally. It is widely recognized as the superior database for bibliometric research in various disciplines, providing excellent literature retrieval and citation analysis services (Cheng et al., 2022b; Zhang et al., 2022; 2023).

In contrast to WoSCC, the freely accessible PubMed database, curated by the United States National Library of Medicine, is updated daily. This database comprises over 34 million citations that are readily accessible via the internet (Cui et al., 2023). The PubMed MeSH (Medical Subject Headings) system uses standardized vocabulary for subject annotations and categorizations of medical literature (Medical Subject Headings Home Page, 2023). MeSH terms can be precisely identified and retrieved in literature databases using a unique identifier code. Consequently, both the WoSCC and PubMed databases provide excellent utility within this study.

We have provided a more comprehensive depiction of our elaborate search approach in Figure 1. 3,123 original articles were retrieved from the WoSCC database by topic searching (TS) and the PubMed database using MeSH and Title/abstract. Subsequently, 3,018 articles were screened based on publication year (2008–2023.05), publication type (articles and reviews), and language (English). Finally, the retrieved files were exported in the form of references.

We conducted distinct searches in the WoSCC and PubMed databases, meticulously removing duplicate papers, and proceeded with subsequent in-depth bibliometric analyses. To ensure the utmost objectivity and comprehensiveness of our research, we scrutinized the annual publication trends of NPART within these two databases. We further prognosticated forthcoming publication trends. Additionally, we provide accurate and comprehensive predictions of NPART research hotspots through keyword analysis on both databases. As for the remaining sections, we analyzed them separately using the WoSCC database.

The grey prediction model GM(1,1) is a commonly used nonlinear prediction method suitable for short-term forecasting (Luo et al., 2020). The primary objective is to analyze the developmental trend of sequence data and establish a grey differential equation model to predict future trends. In this study, we utilized a standard GM(1,1) model implemented through R 4.3.0 software to forecast the annual publication volume of anti-cancer network pharmacology research for the next 4 years.

The records of the retrieved publications were exported to BiblioMetrix, Citespace, and SCImago Graphica for further bibliometric analysis. The BiblioMetrix software package (R4.3.0) captured and extracted bibliographic information about the selected publications, including subjects, authors, keywords, and country distribution (Aria and Cuccurullo, 2017). The H-index measures scientific output and quantitatively assesses the impact of citations. Even if the total number of papers or citations may vary, two individuals with similar H-indices could potentially have similar effects on the scientific field. The H-index considers both the number of publications and the number of citations received by those publications, providing a more comprehensive measure of an individual’s research impact (Hirsch, 2005). The M-Index can be used to compare the effects of academics with those with different careers. The number of citations to a document is partly a measure of its scientific impact (Eyre-Walker and Stoletzki, 2013).

CiteSpace is a Java-based tool that visually analyzes academic literature in specific research fields or disciplines. The software allows for analysis and cluster analysis of citations, keywords, authors, institutions, and countries, to visualize scientific research data of specific fields. The key parameters to measure the network structure are betweenness centrality (BC), modularity, and silhouette (Wu et al., 2022). Firstly, the BC metric measures how well one node connects to other nodes on a network path. As a general rule, nodes with BC values greater than 0.1 are key hubs that bridge between nodes, and are also displayed with bright red rings. Secondly, silhouettes can be used to verify that data clusters are consistent. Thirdly, burst detection was used to detect keywords/references that appeared in high numbers for a specific period of time. Moreover, the dual-mapping overlay method in Citespace consists of two parts: elemental mapping and overlay mapping. Overlay mapping uses generated authority literature data, while elemental mapping uses citation distribution maps (Xu and Feng, 2021). Finally, we can analyze trends and emerging priorities in the research field using a network of keywords and co-cited literature datasets.

SCImago Graphica provides a range of graphical tools, such as journal maps and trend graphs that are useful for visualizing research output and trends at the journal and country levels (Hassan-Montero et al., 2014). By revealing the scientific collaboration relationships, strength, and focus among countries using the national collaboration network, this study aims to shed light on the development trends and focal areas of international academic collaboration.

The alluvial generator in MapEquation visualization is used to visualize the flow and correlation relationships between different categories or variables (Rosvall and Bergstrom, 2010; Blöcker et al., 2023). It highlights significant structural changes to reveal the basis and frontiers of network pharmacology research on anti-cancer in literature co-citation network data. The tool helps to connect structural and functional modifications more intuitively.

This study includes 3,018 articles, with 2,210 in WoSCC and 808 in PubMed. Among the WoSCC are 2,036 articles and 174 reviews authored by 115,951 authors from 6,319 institutions. These articles were published in 492 journals and cited 69,219 times in 1,014 journals. In addition, there are 748 articles and 60 reviews in PubMed, authored by the same group of authors from the same institutions in the same countries. These articles were published in 492 journals and cited 69,219 times in 1,014 journals.

To a certain extent, the number of publications in a field indicates its developmental scale. As depicted in Figure 2, the yearly research output in network pharmacology and anti-cancer research has maintained a consistent growth rate between 2008 and 2022, indicating a positive developmental trend. Notably, there has been a surge in the volume of publications since 2020, with the number of publications reaching 774 in WoSCC and 287 in PubMed for the year 2022, providing evidence of the growing interest of scholars in this field (Supplementary Table S1).

To gain a comprehensive understanding of the pattern of literature growth, we employed the GM(1,1) model to transform the data sequence into a gray differential equation based on the growth trends of publication output data from 2008 to 2022. We derive the predicted publication output for 2023–2026 by exponentiating the accumulated line. The GM(1,1) model’s accuracy using WoSCC and PubMed data is 0.9953 and 0.9965, respectively, indicating high precision. The Annual scientific publications output in the NPART is consistent with the fitting curve of the model, meaning that the literature output in this field will undergo a rapid growth phase over the next 4 years.

To evaluate the performance of a country’s contribution to the NPART field, we first conducted a statistical analysis of the number of articles published by each country. The results showed that China, the United States, and India ranked in the top three. According to Figure 3A, the top three countries in terms of total citation count were China, the United States, and the United Kingdom. Secondly, we used VOSviewer and Scimago Graphica to analyze the collaboration network of the top 30 countries visually. The size of each node was proportional to the number of articles published by that country, with larger nodes indicating a higher number of articles published. Besides, the width of the lines connecting nodes represented the strength of collaboration between countries, with more comprehensive lines indicating closer cooperation. Furthermore, the collaboration network was divided into three clusters, with countries within the same cluster collaborating more closely. Meanwhile, the thickest gray line in different clusters indicated that China and the United States were the leading countries with close collaboration (Figure 3B).

Finally, we can examine the scientific research status of the top 30 countries in the NPART field from the perspective of publications and scientific collaboration. The distribution of publications is uneven among different countries. Regarding the number of publications, China, the United States, Germany, and the United Kingdom are at the forefront of a comprehensive research collaboration network. However, countries like Singapore, Mexico, and Iran are ranked lower.

Supplementary Table S2 provides detailed information on the publications of the top 30 countries, highlighting the variations in collaboration between them. The citation count of a paper reflects its quality. Therefore, China, the United States, and the United Kingdom have conducted in-depth research, resulting in relatively mature outcomes with broad impact. Distributed collaborative research would undoubtedly contribute to the further development of this field if more countries were involved in extensive research.

Through keyword analysis, we classified closely related keywords in anti-cancer research into distinct groups and linked each with a corresponding theme. In fact, the interconnections between these themes are established through the interconnections between the keywords. This process involved a comprehensive summation of the keywords and the identification of the themes using clustering analysis. Based on this, we can make informed inferences regarding this field’s current research frontiers and trends.

This study used Citespace software to generate a keyword co-occurrence visualization map. After combining synonyms and eliminating irrelevant terms, the map contained 679 keywords and 2,793 links. The keywords with a burst phenomenon are represented by central red dots, indicating that the field or theme is rapidly developing or receiving widespread attention (Figure 5A). The appearance of keywords such as “immunotherapy,” “herbal medicine,” “alkaloid,” and “vitamin C″ in the past 3 years suggests the future research direction and hot topics in the NPART field. In addition, we selected the top 20 keywords as core keywords based on their frequency, BC and burst strength (Table 2). Meanwhile, BC is an essential indicator of keyword importance within a network. Keywords with high BC play significant roles in the structure and can be seen as bridges connecting research topics. It is worth noting that “breast cancer” and “mechanism,” as well as “endothelial growth factor” and “angiogenesis,” all have a BC score greater than 0.1. This indicates that these keywords hold prominent positions within the NPART structure. In addition, “tumor necrosis factor” and “anticancer agent” are crucial keywords that cannot be ignored. Their BC scores of 0.09, close to 0.1, also indicate their significant positions within NPART. As a result, identifying these key terms will allow us to better understand which concepts are most relevant to this field.

Moreover, co-occurrence frequency is an essential indicator of research hotspots in this field. We can conclude that keywords such as “apoptosis,” “molecular docking,” and “traditional Chinese medicine” still hold important positions in the citation information presented in Table 2. By conducting burst analysis, we can observe that emerging research frontiers often experience explosive growth quickly before either evolving into the following popular topic or disappearing altogether. Thus, by analyzing burst trends over time, we can gain insights into how research topics evolve within this field and identify potential areas for future investigation. Over the past 5 years, NPART has exhibited a trend towards diversification and has attracted broad attention in several fields such as “invitro”, “prediction”, “database”, “combination therapy”, and “natural product.” At the same time, keywords like “drug discovery,” “systems pharmacology,” and “systems biology” continue to be areas of future research interest. These findings suggest that the field of NPART is continually evolving and expanding into new frontiers. As researchers continue to explore these emerging topics while also investigating established ones, we can expect future advancements in this field.

The visualization of keyword clustering in Figure 5B provides a clear overview of the topics studied in this field and their interrelationships. At the same time, we use a timeline to represent the occurrence of each keyword. This is particularly helpful in understanding the evolution of these themes. Furthermore, the silhouette and module scores confirm the validity of the clustering results. The connection between “combination therapy” (Nelander et al., 2008) in the #2 topic and “anticancer drug” (Kong et al., 2016) in the #5 topic suggests that these two concepts are related, as they both appeared during a specific period (Figure 5C). This finding may be helpful for researchers who are interested in developing new anticancer therapies. Additionally, their temporal continuity makes it possible to distinguish classical topics from emerging ones. Standard topics such as “#0 database”, “#1 discovery”, “#2 network pharmacology”, “#3 apoptosis”, and “#4 suppress” show good temporal continuity, meaning they have been studied for an extended period with sustained interest from scholars. In contrast, emerging topics like “#5 anticancer agents”, “#6 oxidative stress”, and “#7 nf kappa b” have recently attracted attention but require more research to establish themselves within the field.

In addition, we can identify current research hotspots such as “Tumor Microenvironment”, “Traditional Chinese Medicine”, “Plant Extracts”, “Molecular Docking Simulation”, and “Anti-neoplastic Agents” using heatmaps to visualize frequency accumulation across all documents for each keyword (left panel), as well as occurrence counts for different years (right panel). We conducted an annual keyword analysis of relevant literature in the PubMed database to predict future research trends. This serves both as a supplement and a validation to the keyword hotspot analysis results from WoSCC database. As a result, the above keywords could become hotspots and trends for future research in this field.

For literature to be cited together, it must possess similar content. When two or more studies subsequently cite the same articles, a co-citation relationship is established between them. Co-citation reference analysis is a method for measuring the relationship between reference literature. Therefore, this method can uncover research themes and evolutionary trends within a particular field by incorporating cluster analysis.

Figure 7A displays the top 10 most cited articles and utilizes CiteSpace to present their co-citation network. In Figure 7A, nodes with high BC are marked in purple outer rings, while in Figure 7B, nodes with citation bursts are marked in red circles. Additionally, we identified the top 25 articles related to NPART with the most robust citation bursts shown in Supplementary Figure S2. These articles were classified into 11 categories (Supplementary Table S4), and their degree of correlation served as the basis for cluster classification (Figure 7C). The most significant cluster was “#0 molecular docking”. In contrast, the earliest research cluster was “#5 clinical trial enrichment”. In 2023, it was discovered that there was a high frequency of connections among different groups, such as “#0 molecular docking”, “#9 calycosin”, “#1 molecular mechanism”, and “#3 anticancer agents” (Figure 7D).

In addition, we generated a literature citation network using Citespace from 2018 to May 2023 to explore the key literature on NPART. Subsequently, we imported the network information files for each year into the streamflow generator program and obtained a streamflow map of landmark literature following filtering and layout adjustments. Figure 8 shows four highly acclaimed milestone articles with a streamflow continuity of 4 years between 2018–2021 and 2020–2023.

Between 2018–2021, four papers were published on the topic, with the first being Kim et al., 2016 publication “PubChem Substance; Compound databases” (Kim et al., 2016). This literature delves into PubChem as a public chemical substance and biological activity information database, offering chemical information resources and search interfaces, supporting data submission and programming access, and featuring derivative resources such as PubChem3D and PubChemRDF. In network pharmacology, these resources are utilized for drug discovery and design. Secondly, Kanehisa_M introduced KEGG as an encyclopedia of genes and genomes, encompassing the KO database, networks of molecular interactions, reactions and relations, BRITE hierarchy, and KEGG modules (Kanehisa et al., 2017). In fact, KEGG has proven to be a valuable tool for predicting drug action mechanisms, identifying new drug targets, and designing novel drug molecules. Through enhancements to the DISEASE and DRUG databases, KEGG can integrate diseases and drugs with molecular networks. Thirdly, Szklarczyk_D developed the STRING and STITCH5 databases for NPART development. STRING is a valuable tool for predicting drug action mechanisms and identifying new targets (Szklarczyk et al., 2017). By analyzing the interactions between drugs and proteins, it is possible to predict the mode of action of drugs and potential side effects. This makes STRING an essential resource for drug discovery and development. Furthermore, the STITCH5 database is also a valuable resource that integrates multiple data sources, including interaction information between proteins and small molecules, covering 430,000 chemical substances (Szklarczyk et al., 2016). It provides a new network view that allows users to view the binding affinity of chemical substances in the interaction network, thereby facilitating a quick understanding of the potential impact of chemical substances on their interaction partners. This makes STITCH5 a valuable tool for drug discovery and design. In summary, these tools discussed above are of immense significance for network pharmacology research, particularly in anti-cancer research. These articles represent classic works of original network pharmacology research and serve as the foundation of this field.

In addition, the methods and tools proposed in the four critical articles from 2020 to 2023 have gained widespread attention and recognition for their frontier nature and innovativeness in the NPART. These articles represent important research achievements in this domain, which are significant for researchers and clinicians seeking insights into the network pharmacological mechanisms underlying anti-cancer treatment strategies. Firstly, Zhou_ZC has pointed out that network pharmacology can be employed to elucidate the multi-target characteristics of traditional Chinese medicine, particularly for treating complex diseases (Zhou et al., 2020). Furthermore, network pharmacology offers a novel research approach to investigating traditional Chinese medicine. Secondly, Wang_YX established the Therapeutic Target Database, which encompasses the regulatory effects of microRNAs and transcription factors on targets, protein-target interactions, patented drugs, and their corresponding targets (Wang et al., 2019). These data are derived from other validated, clinically tested, literature-reported target collections. Thirdly, OMIM.org is a website constructed by Amberger_JS to integrate the knowledge of phenotype-gene relationships. The website documents have known genetic diseases and are a primary repository for information on gene-phenotype associations (Amberger et al., 2019). Lastly, the Swiss Target Prediction tool aims to predict protein targets of small molecules (Daina et al., 2019). The device has a collection of 376,342 compounds with known experimental activity and searches for similar 2D and 3D molecules to predict 3,068 large molecule targets. The latest version has a predictive performance of over 70% and can correctly predict at least one human prey in the top 15 predictions. The tool has been applied in anti-cancer research to predict the interactions between small molecules and tumor-related proteins and guide drug design.

In summary, references with high BC, high citation counts, and citation durations of 4 years are essential to advance our understanding of NPART. In addition, the evolutionary relationship between the cited literature clusters suggests that “#0 molecular docking”, “#1 molecular mechanism”, “#4 network medicine”, and “#9 calycosin” may become hotspots and frontiers.

We also analyzed the overall evolutionary trends in this field. From keywords perspective, network pharmacology and cancer have emerged as core and foundational research components in the entire domain. Notably, keywords such as, traditional Chinese medicine, and anti-cancer research have gained prominence in recent years, indicating the significant research potential in these areas and encouraging scholars to explore them further. Regarding the clustering of keywords in the articles, clusters “#0 databases” and “#5 anticancer agents,” were among the earlier and larger keyword clusters (Figure 5B). The emerging keywords in this field include traditional Chinese medicine, tumor microenvironment, and Phosphatidylinositol 3-Kinases (Figure 5D). In this perspective, scholars have been studying the connection between network pharmacology and anti-cancer and its microenvironment since 2016, and the popularity of these keywords has persisted. Recent studies have demonstrated that ginger regulates the proliferation, migration, and apoptosis of triple-negative breast cancer cells by modulating signaling pathways such as TNF, IL-17, FOXO, MAPK, PI3K/AKT, and others (Luo et al., 2023). In addition, tumor radiotherapy often leads to the development of radiation-induced oral mucositis in patients. A formulation consisting of extracts from Lonicera japonica, Paeonia lactiflora, and Cornus officinalis has been employed for the treatment of radiation-induced oral mucositis, targeting multiple pathways, including Toll-like receptor signaling and TNF signaling cascades (Sheng et al., 2023). In summary, Chinese herbal medicine, as an integral part of traditional medicine, demonstrates promising anti-cancer activity and therapeutic potential in cancer treatment. Network pharmacology research enables us to understand Chinese herbal medicine’s active constituents, targets, and mechanisms of action, thereby providing a scientific basis for its application in cancer therapy.

By analyzing the literature citation count, we can identify research hotspots and cutting-edge fields that have garnered widespread attention and citation in the academic community (Wang J. et al., 2023). The 11 clusters identified from the literature citations cover areas such as traditional Chinese medicine, tumor immunology, and clinical trials, indicating that this field has attracted extensive attention from interdisciplinary scholars. The academic influence of these clusters extends beyond individual disciplines, fostering interdisciplinary collaborations and knowledge exchange and driving cross-disciplinary integration and innovation (Tang et al., 2023). The compound Chinese medicine Xihuangwan exerts inhibitory effects on the progression of breast cancer by targeting multiple proteins involved in the inflammatory microenvironment signaling pathways, including FOS, MYC, JUN, PPARG, MMP9, PTGS2, and SERPINE1 (Wu et al., 2020). By applying a network pharmacology approach, Guo et al. identified five components of CKI that may target 187 esophageal cancer-related genes and inhibit genes in cancer cell-associated pathways. The active ingredients identified in the compound injection include quercetin, flavonoids, nobiletin, lignans, and β-sitosterol. These active ingredients have shown potential in treating esophageal cancer (Guo et al., 2023).

In particular, the cluster analysis in 2023 indicates a close relationship and interdependence between cluster #0 (molecular docking), cluster#1 (molecular mechanism), cluster #3 (anticancer agents), and cluster #9 (calycosin), reflecting the current research trends and development directions of NPART. Calycosin, a primary active ingredient isolated from Astragalus membranaceous (Huangqi), has been shown to exhibit significant therapeutic effects on colon cancer by inhibiting the ERK1/2 signaling pathway (Hu et al., 2023). It is worth noting that two distinguished traditional Chinese medicine physicians have invented a core formula comprising Astragalus membranaceous (Huangqi), Salvia miltiorrhiza (Danshen), Carthamus tinctorius seeds (Suhonghuazi), Hedyotis diffusa (Baihuasheshecao), Scutellaria barbata (Banzhilian), and Curcuma zedoaria (Ezhu). This formula may exert anticancer effects by targeting 14 liver cancer-related genes, primarily inhibiting inflammation and angiogenesis and enhancing the immune response (Bochuan et al., 2023).

In summary, we applied keyword and literature co-citation cluster analysis to reveal that the current research hotspots mainly focus on databases, herbal medicines, molecular docking, and tumor immune microenvironment regulation. Furthermore, the latest publications indicate that Chinese herbal medicines are developing into compelling cutting-edge research hotspots for fundamental research and clinical application. These research directions provide new perspectives and opportunities for the further development of Chinese herbal medicine anti-tumor therapy.

Although network pharmacology holds significant value in cancer treatment, it still faces challenges regarding data quality, algorithm selection, and reproducibility. From the perspective of application value, network pharmacology demonstrates tremendous potential in anticancer therapy. First of all, network pharmacology can identify key molecular targets and pathways involved in tumor development and progression by integrating various data sources such as genomics, proteomics, and metabolomics. Secondly, it can also predict the efficacy and toxicity of candidate drugs and optimize drug combinations for personalized treatment. Lastly, network pharmacology provides a systems-level understanding of drug interactions with targets, which is crucial for developing safer and more effective anticancer therapies.

However, some challenges need to be addressed when applying network pharmacology to anticancer treatment:

1. The integration and analysis of large-scale omics data require advanced computational and statistical methods, which can pose technical difficulties for researchers.

In conclusion, network pharmacology holds tremendous potential in anticancer treatment, but some challenges must be addressed. Future research should focus on developing more advanced computational and experimental methods, improving data quality and availability, and exploring personalized and precision medicine strategies.

There are several limitations to this study despite these interesting findings. First, the data for this study were obtained from WoSCC and PubMed databases analyzed separately. Additionally, the Scopus database was not used in this study due to its restricted availability. Secondly, the WoSCC literature contains comprehensive information and is often considered the most appropriate bibliometric database (Cheng et al., 2022a; Chen et al., 2022; Zhong and Lin, 2022). However, the PubMed database lacked information on cited literature and could not be analyzed for literature co-citation. Finally, due to the limitations of these bibliometric software, the specific parameter settings for the relevant analyses are somewhat subjective. This means that they may miss potentially high-impact studies.