Traditional Chinese Medicine (TCM) is one of the greatest treasures of Chinese culture, has a long history of use in East and Southeast Asia, and has been widely used since ancient civilization. Through continuous development and innovation, physicians have selected the essence and discarded the dross of TCM. As a result, TCM has become one of the main forms of alternative medicine in East Asia, North America, and Europe. TCMs use therapeutic herbs to treat diseases according to the combinatorial principle of “King, Vassal, Assistant, and Delivery servant” based on a patient’s syndrome and to restore balance of life and body functions (Yi and Chang, 2004; Qiu, 2007). Each prescribed combination of these herbs is referred to as a TCM preparation, or TCM prescription, for example the LiuWeiDiHuangWan (LWDHW) pill. Traditionally, the discovery of most new drugs is focused on identifying or designing a pharmacologically effective agent that specifically interacts with a single target. During the past 30 years, such an approach has generated highly successful drugs. However, drugs acting on individual molecular targets usually exert unsatisfying therapeutic effects or have toxicity when used to treat certain diseases, such as diabetes, inflammation, and cancer (Kola and Landis, 2004). Given the major bottleneck in drug discovery, drug research, and development have gradually shifted from a “one-target, one-drug” mode to a “network-target, multiple-component-therapeutics” mode (Li, 2011; Li et al., 2011, 2014b). To treat disease, the holistic philosophy of TCM shares key concepts with the new mode of drug discovery. Therefore, making full use of TCM is of great importance for drug research. In recent years, TCM has gradually garnered interest thanks to its low toxicity and therapeutic effects (Chan, 1995; Corson and Crews, 2007; Qiu, 2007). For example, Ganoderma lucidum, termed “LingZhi” in China, has been used as a health-preserving and therapeutic agent. Previous studies have shown that it possesses medicinal properties, including anti-cancer, anti-diabetic, anti-hepatotoxic, and immunomodulatory effects (Paterson, 2006; Boh et al., 2007, Boh, 2013; Ma et al., 2015). Panax ginseng is another TCM that has been used therapeutically for more than 2,000 years to treat such diseases as cardiovascular disease, Alzheimer’s disease, and diabetes (Lee et al., 2008; Kim, 2012; Yuan et al., 2012; Chan et al., 2013). Unlike modern drugs discovered by targeting a specific protein, however, the understanding of the molecular basis of TCM remains limited, and research into modern TCM theory has lagged, which has slowed down the development of novel TCMs (Corson and Crews, 2007). TCM studies must keep pace with the rapid development of modern science to remain relevant. Therefore, a scientific and effective assessment system needs to be established, which will be key for studying and making full use of TCMs.

Given the rapid progress in bioinformatics, systems biology, and polypharmacology, network-based drug discovery is considered to be a promising approach for cost-effective drug development. Systems biology examines biological systems by systematically perturbing them; monitoring the gene, protein, and informational pathway responses; integrating these data; and, ultimately, formulating mathematical models to describe the structure of the system and its response to individual perturbations (Ideker et al., 2001). Based on a systems biology approach, the concept of network pharmacology was first proposed by Li et al. (2014b). Because network pharmacology can provide a full or partial understanding of the principles of network theory and systems biology, it has been considered the next paradigm in drug discovery (Hopkins, 2008). Furthermore, the network pharmacology approach has been used to study “compound-proteins/genes-disease” pathways, which are capable of describing complexities among biological systems, drugs, and diseases from a network perspective, sharing a similar holistic philosophy as TCM. Applications of systems biology methods to determine the pharmacological action, mechanism of action, and safety of TCMs are invaluable for modern research and development of TCM. Thus, a new interdisciplinary method termed TCM network pharmacology has been proposed (Li and Zhang, 2013; Liang et al., 2014a; Li et al., 2014b), which has initiated a new research paradigm for transforming TCM from an experience-based to evidence-based medicine. In this work, we first summarized the currently widely used databases and tools for TCM network pharmacology research. Second, we concentrated on the different applications of network pharmacology to TCM research, including TCM recipes, target prediction, and network toxicology. Third, we compared different TCM databases based on their basic properties and search results (Figure 1).

Traditional Chinese Medicine preparation (in other words, TCM prescription) is the main form of TCM, and it is produced under the guidance of TCM syndrome differentiation and treatment, which is highly scientific. The traditional “King, Vassal, Assistant, and Delivery servant” combination rule of TCM herbal formulas contains the rich and profound scientific connotation of TCM theory. Thus, the ability to explain the compatibility of TCM is one of the most challenging tasks for TCM research. Li et al. (2010) proposed a method named the Distance-Based Mutual Information Model (DMIM) to identify useful relationships that exist among herbs in numerous herbal formulas. DMIM combines mutual information entropy and “between-herb-distance” to score herb interactions and construct an herb network. It achieves a good balance among the herb’s frequency, independence, and distance in herbal formulas when used to retrieve herb pairs. They used 3,865 collateral-related herbal formulas to construct an herb network and illustrated the traditional herbal pairing and compatibility. LWDHW was used as an example to identify possible compatibility mechanisms, and it was found that LWDHW-treated disease shows high phenotype similarity and that certain “co-modules” are enriched in cancer pathways and neuro-endocrine-immune pathways, which may be the mechanism of action by which the same LWDHW formula treats different diseases. Zhang et al. (2008, 2017) focused on the five flavors of Chinese medicinal properties and constructed Bayesian network models of bitterness, pungent flavor, and sugariness. The five flavors of some TCMs were predicted by these established models, providing crucial support for research on TCM compatibility.

Understanding the molecular basis of TCM is crucial for the modernization of TCM, which not only would provide full theoretical support for TCM research but also would increase acceptance of TCM worldwide. In recent years, researchers have done a lot of work to study the molecular mechanisms of TCMs. We proposed the TCM-Mesh system (Zhang et al., 2017), which was designed as an integration of a database and a data-mining system for network pharmacology analysis of TCM preparations. We used TCM-Mesh to identify candidate targets for ginseng and LWDHW based on existing clues (e.g., molecular structures, bioactivity) from curated databases along with a pharmacology approach. The results showed that dammarane-type tetracyclic triterpenoids from ginseng (i.e., ginsenoside Rg1, ginsenoside Re, and ginsenoside Rb1) might be used to treat Alzheimer’s disease, hypertension, and atherosclerosis by targeting tumor necrosis factor (TNF), nitric oxide synthase 3 (NOS3), and AKT serine/threonine kinase 1 (AKT1). Chemical constituents from LWDHW (i.e., loganin, paeonol) can be used to treat diabetes, diabetic nephropathy, and diabetic retinopathy by targeting intercellular adhesion molecule 1 and connective tissue growth factor. Li et al. (Liang et al., 2014b) established a novel pharmacology approach based on a network pharmacology approach to analyze the traditional herbal formulas for LWDHW. The authors found that the compounds of LWDHW may play an important role in esophagitis and colon cancer by regulating the expression of C-C chemokine receptor 2 (CCR2), estrogen receptor 1 (ESR1), peroxisome proliferator-activated receptor gamma (PPARG), and retinoic acid receptor alpha (RARA). They also selected a XinAn medical family’s anti-rheumatoid arthritis (RA) herbal formula “QingLuoYin” (QLY) as an example (Zhang et al., 2013), which consisted of four herbs: KuShen (Sophora flavescens), QingFengTeng (Sinomenium acutum), HuangBai (Phellodendron chinensis), and BiXie (Dioscorea collettii). Some anti-angiogenic and anti-inflammatory active ingredients, such as kurarinone, matrine, sinomenine, berberine, and diosgenin, were identified among the 235 ingredients of QLY. In addition, the synergistic effects of the major ingredients were identified in QLY, such as matrine and sinomenine, which may have been derived from the feedback loop and compensatory mechanisms by targeting TNF- and vascular endothelial growth factor-induced signaling pathways involved in RA. According to the molecular structures of proteins and the corresponding targets of 1,401 drugs approved by the U.S. Food and Drug Administration, Wu et al. (2011) constructed a target prediction model by using the “Random Forest” algorithm. The authors used the compounds of “fuzi” (Aconiti Lateralis Radix Praeparata) to predict the targets and construct a multi-target network. The study found that the 22 compounds of “fuzi” have been used to predict several targets that reflect the characteristics of TCM. In addition, Li et al. (An and Feng, 2015) studied the antioxidant effects of ZhiZiDaHuang Decoction (ZZDHT) based on a network pharmacology approach. The authors found that ZZDHT may exert antioxidant effects to regulate reactive oxygen species, thereby treating alcoholic liver disease by targeting cytochrome P450 2E1 (CYP2E1), xanthine dehydrogenase (XDH), nitric oxide synthase 2 (NOS2), and prostaglandin-endoperoxide synthase 2 (PTGS2).

Network toxicology is based on an understanding of the “toxicity-(side effect)-gene-target-drug” interaction network, which utilizes network analysis to speculate and estimate the toxicity and side effects of drugs. Liu et al. (2012) proposed that technologies that feature rapid preparation, high-throughput screening, toxic components exclusion, and biochip along with the drug-target network are of great importance to TCM research on active ingredient screening, toxic components exclusion, and molecular mechanism, which could increase the safety of TCMs. Fan et al. (2011) proposed the concept and framework of network toxicology of TCM. The related tools and technologies were briefly introduced, and the prospects for network toxicology of TCM were forecasted. In their work, they used network pharmacology methods to reconstruct the network of “compound-protein/gene-toxicity” to identify toxic substances and predict the toxic side effects of known compounds, which provides valuable information for understanding the toxic mechanisms. Currently, the related databases for the study of network toxicology includes CTD (Davis et al., 2011), TOXNET (Wexler, 2001), and the National Toxicology Program. Furthermore, additional foresting toxicity software, such as TOPKAT, HazardExpert, and DEREK (Wolfgang and Johnson, 2002), are available for TCM network toxicology studies.

As discussed, various databases are used for TCM analysis. However, network analysis of TCM is limited by several aspects of the currently available databases (Figure 3, current statistics on various databases can be obtained from the homepage of each database).

For example, HIT is a comprehensive TCM database that supports many different input formats. It is limited, however, in that the total information from HIT for TCM analysis is not sufficient, as it only contains about 586 compounds and 1,301 targets. Thus, the data that can be collected from the HIT are limited. TCMSP is a TCM database that captures the relationships among drugs, targets, and diseases. It allows users to use many input formats; however, TCMSP should be more concise as the database has a large amount of redundancies when searching for herbs, which must be trimmed to be improved. TCMID is another comprehensive database that provides information and bridges the gap between TCM and modern life science. It collects TCM-related information from the TCM@Taiwan database and the literature, including prescription of TCM, and also offers a module for network visualization. Although TCMID is available for multiple inputs, including compound name, STITCH database ID, and CAS number, when searching for ingredients, it is not sufficiently comprehensive, and many compounds are searched without a related network in TCMID. The newly launched TCM-Mesh is a more comprehensive database that not only contains the data on network pharmacology, including the “herb-compound-target-disease” network, but also includes data on TCM side effects and toxicity. As a result, TCM-Mesh offers a more holistic perspective for those conducting TCM research. Problems in TCM-Mesh remain to be resolved, however; for example, the web service is limited and many functions need to be improved, as the CAS number query is not supported. In addition, detailed descriptions about the compounds, genes, and diseases are lacking.

We used prescriptions for LWDHW and ZZDHT to assess TCM-Mesh and randomly selected 10 well-known herbs to test different TCM databases for systematic assessment (Table 3).

As mentioned earlier, Li et al. (2010) found that the compounds of LWDHW may play an indispensable role in esophagitis and colon cancer by regulating the expression of CCR2, ESR1, PPARG, and RARA. To validate the effects of LWDHW on esophagitis and colon cancer, we used TCM-Mesh. We found 11 candidate targets (including PPARG) for colon cancer and four potential targets for esophagitis (Table 4), all of which were validated by the literature.

As discussed in Section 3.2, “Target prediction of TCM,” ZZDHT decoction has been applied to treat alcoholic liver disease by targeting CYP2E1, XDH, NOS2, and PTGS2. We used TCM-Mesh to validate the effects of ZZDHT, leading to the identification of 12 candidate targets (including CYP2E1) for alcoholic liver disease (Table 5), all of which were supported by the literature.

To further compare the databases, we selected 10 well-known herbs to randomly evaluate the TCM databases: G. lucidum, ginseng, Codonopsis pilosula, Astragalus membranaceus, Chinese yam, pseudo-ginseng, Polygonum multiflorum, Radix Angelicae dahuricae, Coptis chinensis, and Cordyceps sinensis. We used each herb to search the different databases and compared the search results (Table 6).

We compared the results from different databases and found that although the compound data recorded in TCM-Mesh were abundant, the data on the ingredients of the herb were limited. For each herb, the number of ingredients obtained from TCM-Mesh was lower than the number obtained from TCMID and TCMSP. We also found that the number of ingredients collected from TCMSP was much higher than that collected from the other two databases, in accordance with the fact that the database had a large amount of redundancies when searching for herbs. Additionally, we discovered higher completeness of TCM-Mesh in the ingredient-target associations and target-disease associations. For many herbs, the search results produced more hits in ingredient-target associations and target-disease associations in TCM-Mesh, although fewer ingredients were collected. Furthermore, a search for P. multiflorum in TCMSP produced no hits, revealing the limited herb data in TCMSP. The TCM database must keep pace with the development of TCM research to support the modernization of TCM. Therefore, a more comprehensive and complete database with powerful web services is necessary.

In recent years, several databases have been proposed, many of which share similar functions although they have different data sources. For most databases, the patterns of input differ from database to database. For example, the herb “Dried Tangerine Peel” in TCMSP must be searched by “chenpi” or “Citrus reticulata,” whereas in TCMID, it must be searched by “chen pi” or “Citri reticulatae pericarpium.” In addition, many compounds and proteins have different aliases in different databases (e.g., compounds: chemical name, CID number, STITCH ID, CAS number, PubChem CID, EC number, UNII; proteins: protein name, gene symbol, node ID, target ID, target drugbank ID, uniprot ID). The uniform format of input and output not only would make the current databases more concise, but also would enable researchers to take full advantage of the database resources. Thus, a standard transforming platform is needed. Such a platform should fully understand the data structure and content of various databases. Furthermore, it should support the translation among different formats of herbs, compounds, or proteins to bridge the gap between current databases for TCM research. Currently, the output of previous databases can be directly used as the input for another database. Most databases have associated web services; thus, the transforming platform that could transfer data from one database to another, can be presented to users as web services that are linked to different databases or to a browser plug-in. In addition to the unification of databases, the integration of diverse databases is also of great importance. Currently, more than 10,000 herbs used in more than 100,000 herbal formulas have been recorded in TCM (Qiu, 2007), posing a huge challenge for researchers who want to better understand the efficacy and the safety of the TCM. It is difficult for a single database to resolve all these problems as a result of the incompleteness of data. Each database has its own advantages regarding the specificity of its data and its algorithm. Therefore, the full integration of available resources will facilitate the production of an even more powerful and comprehensive database system, which undoubtedly will more effectively promote the modernization of TCM.

TCM aims to restore whole-body balance in patients by using a herbal formula (Li et al., 2007), and herbal prescriptions are usually composed of two or more medical herbs in a certain proportion. For example, LWDHW consists of six herbs (Zhao et al., 2007; Xie et al., 2008): Rehmannia glutinosa libosch., Cornus officinalis Sieb., Dioscorea oppositifolia L., Paeonia ostii, Alisma orientale Juz, and Poria cocos (Schw.) Wolf, with a dose proportion of 8:4:4:3:3:3 (Wu et al., 2007; Wang et al., 2010). For current studies on TCM formulas, researchers focus on the chemical composition rather than on the proportion of each compound of TCM. Therefore, the influence of the compound content on the effect of the TCM is often ignored. The weight of the content of each compound should be taken into consideration in future studies to better understand the pharmacological effects of the TCM.

The significant involvement of the gut microbiota in human health and disease suggests that manipulation of commensal microbial composition through combinations of probiotics, antibiotics, and prebiotics could be a novel therapeutic approach. Previous studies have shown that many TCMs can be used as agents to prevent gut dysbiosis (Chang et al., 2015; Zhou et al., 2016). Therefore, a more comprehensive database about TCM and gut microbiota is needed, which should not only include information on the interaction between the TCM and the gut microbiota but also integrate respective advantages of TCM databases and microbiome databases. This integration will be conducive to accelerating the internationalization of TCM and will allow researchers to fully understand the efficacy of TCM from a holistic perspective. Research in the area of TCM network pharmacology is still in its infancy and remains to be developed. However, with data accumulation on TCM and clinical research and interaction of various methods of analysis and experiment, researchers can obtain more substantive and authentic information. This information is conducive to the modernization and internationalization of TCM and may offer critical technological support for drug development, clinical diagnosis, and personalized medicine.