We used a semantic tool for knowledge base management, BioXM (Biomax Information AG, Munchen, Germany), which was developed for object-oriented semantic integration. In this approach, semantically identical objects and the associations between them are identified and mapped based on data and descriptive meta-information [10]. This primarily involves the mapping of biological objects and descriptive data from literature to common objects like genes, diseases, or chemical. Associations between entities are mapped as relations (e.g., chemical X is associated with disease A) and object relation information is contextually structured (e.g., gene B is regulated negatively/positively by chemical X). Using objects as nodes and relations as edges, a “semantic network” can be generated that provides information on the connections between participating objects. Various elementary concepts, such as elements, relations, and annotations, are defined in Fig. 1 and the study workflow is illustrated in Fig. 2. We modified this complex semantic network to detect connections, extract patterns, and answer complex questions, as is shown in Fig. 3. To access the data easily, we used the BioXM tool, which provides visual browsing of the knowledge network (Fig. 3), as well as data mapping and query wizards (Fig. 4) that automatically adapt to the changes of a given data model. This system enabled us to create knowledge networks with flexible workflows for handling data and information types.

Pathway Studio software 9.0 (Elsevier, Atlanta, GA, USA) was used to collect information on biological processes and diseases related to kimchi, based on information in the Korean Food Standard Reference (KFSR). Pathway Studio software allows the automatic extraction of regulatory and physical interactions from MEDLINE abstracts using a natural-language-processing technology called MedScan [11]. Interactions extracted by MedScan, which contain a formalized set of relationships, are imported into the Pathway Studio database and analyzed further using data-mining tools for knowledge inference [12]. Since MedScan keeps the reference of the original article with information on the extracted interaction, it also allows a quick review of extracted facts and identification of relevant publications. Thus, it facilitated our pathway reconstruction by selecting the appropriate interactions. All interactions found were manually curated, and only validated interactions were included in the pathways.

Elements and relations were maintained and linked by relation and object. This semantic model enabled elements to be mapped to known elements. Relations between biological data are used to efficiently extract interactions of interest. The semantic modeling in this study was configured in the generic BioXM knowledge management environment to create a knowledge base for this translational systems biology approach. We constructed semantic networks for the discovery of a network biomarker using BioXM software (Biomax Information AG), which is a customizable knowledge base management tool for scientific data. The graphical data model is shown in Fig. 1. As shown in Table 1, we created elements such as “Foods,” “Foods Group,” “Food Name,” “Metabolites,” and nine other elements to describe the available biological knowledge. Information on the interconnection of elements is captured in relations such as “is related to” for “Foods” or “regulates” for a metabolite that regulates expression of a gene. This network consisted of 13 elements and 20 relations. The semantic network provided information on connections between included objects instantly. Therefore, this integrated semantic data model was used to create new knowledge networks with flexible workflows.

This semantic model enabled us to integrate existing public databases and experimental data derived from the literature. To populate the knowledge base with data from public databases, we manually generated mappings in import-templates. A graphical wizard for import-template generation provided a selection of possible import options and mappings.

The United States Department of Agriculture (USDA) food and nutrient databases provide the basic infrastructure for food and nutrition research, nutrition monitoring, policy, and dietary practice. The databases date back to 1892 and they are unique, as the only databases available in the public domain that perform these functions. The Nutrient Data Laboratory develops and maintains the USDA Standard Reference and related data products [13]. The Standard Reference is the major source of food composition data in the United States. It provides the foundation for most food composition databases used in food policy, research, dietary practice, and nutrition monitoring. The USDA databases were used to construct the KFSR. This product is the basis of numerous other applications that address nutrients in foods. The database contains information on agricultural commodities, as well as data on formulated and processed foods. Currently, the database contains values for up to 140 different food components, including nitrogen (protein), total fat, carbohydrates, and moisture, as well as individual fatty acids, amino acids, vitamins, and minerals in >7,500 foods. We have added five traditional Korean foods, i.e., kimchi, bulgogi, bibimbap, doenjang, and gochujang. To construct a useful KFSR, we integrated 11 public databases as shown in Table 2, with a total of 26,848,859 records.

We found a total 4,351 genes strongly associated with kimchi's secondary metabolites using the KFSR, as shown in Fig. 3. First, 1,237 kimchi metabolites were extracted form >7,500 foods in KFSR and 4,351 genes were associated with 1,237 kimchi metabolites. Through the semantically integrated datasets, we discovered genes that were associated with kimchi metabolites. This network is shown in Fig. 4. We used the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway to classify genes by metabolic function. The genes in Table 3 are classified into six categories, and 309 of these genes were found to be involved in lipid metabolism. At first, we initiated a cardiovascular disease network analysis related to kimchi using 30 genes. Of these, 27 were directly connected by interactions such as regulation, promoter binding, direct regulation, protein modification, and miRNA regulation. Genes like ADCY5, ADCY1, ADCY2, GNAS, CTNNB1, EDN1, and RAC1 were the most highly connected nodes with node degrees of ≥10 (p < 0.001) (Fig. 5A). Adenyl cyclases (ADCYs) are critical regulators of metabolic and cardiovascular function, and ADCY5 is increased in hearts with pressure-overloaded left ventricular hypertrophy [14,15]. In addition, guanine nucleotide-binding protein Gs (GNAS) overexpression results in increased sensitivity to apoptotic stimulation and cardiomyopathy [16], as shown in Fig. 4A. In addition, GNAS leads to the activation of β-catenin which is a multifunctional protein coded by the CNNB1 gene. Although there has long been speculation on the association between kimchi and heart disease, Lee et al. [16] demonstrated that kimchi may not adversely affect blood pressure and cardiac function.

Kimchi is effective in enhancing immune function [3,18,19]; however, the scientific evidence to support this claim has been insufficient. In this work, we performed an immune disease network analysis related to kimchi, based on public information and using 39 selected genes. Twenty-three genes were directly connected to each other by interactions such as regulation, promoter binding, direct regulation, protein modification, and miRNA regulation. In particular, PTPRC, LCK, JAK3, ZAP70, and VEGFA were the most highly connected nodes with node degrees of ≥7 (p < 0.001) (Fig. 5B). Specifically, PTPRC and JAK3 were found to regulate innate immune responses [20,21,22]. This analytic information database, the KFSR, may suggest ways to analyze and more effectively use existing scientific data.