neo4jsbml: import systems biology markup language data into the graph database Neo4j

Introduction

Genome-Scale Metabolic Models (GEMs) of biological systems are commonly represented using the Systems Biology Markup Language (SBML) format. This format is actively maintained and updated, especially, by adding new features to meet new needs. Using a standard format facilitates interoperability repeatability, and reproducibility. Several standards have emerged in systems biology (Shin et al., 2023). The CellML standard stores computer-based mathematical models (Clerx et al., 2020) and the Simulation Experiment Description Markup Language (SED-ML) (Waltemath et al., 2011) enables the reproduction of simulation experiments. The use of a standard format to represent data enables the creation of a rich software environment such as COPASI (Hoops et al., 2006), Tellurium (Choi et al., 2018) and MASSpy (Haiman et al., 2021) for the construction, simulation, and visualization of dynamic metabolic models. A wide range of issues can be addressed in this way, including the construction of metabolic pathways (Shen et al., 2020), study of secondary metabolism in bacteria (Qiu, Yang & Zeng, 2023) and representation of cell behavior and interactions in cancer (Kazerouni et al., 2020). To represent these types of biological processes, the SBML format encodes several nested components and their interactions in an XML-based document. The standard represents complex systems by organizing the data as components. To illustrate, biological systems can contain thousands of metabolites and reactions, leading to a high level of complexity in their interactions (Hucka et al., 2019). The SBML Level 3 standard is well-suited for describing reaction-based models. To represent different types of models, such as constraint-based, logical network, and rule-based models, the SBML standard has been enhanced using different packages. As described in their specifications, these packages add components or properties to the model.

Neo4j is a powerful graph database that enables the storage, querying, and analysis of large amounts of data. Unlike relational databases, Neo4j leverages the power of graph theory, representing data as nodes or relationships which embed properties to store additional information. The graph-based structure is well-suited to tackle intricate relationships and decipher complex problems, such as social networks and network analysis. The graph query language, named Cypher, allows querying the graph to retrieve connected data and perform create, read, update, or delete operations on the database. Neo4j ensures data integrity and consistency by respecting the ACID principle (Meier & Kaufmann, 2019). Specific capabilities can be added to the Neo4j database through a software extension, a plugin, enabling users to customize and enhance its features for dedicated applications. The Neo4j database comes with a broad ecosystem which includes tools like Arrows for designing the database structure, numerous drivers for communicating with the database, building queries intuitively with Cypher and Bloom for visualizing data.

With Neo4j, metabolic models can be represented as a network of nodes and relationships, where each node represents a biological or supporting object within a component, such as a metabolite, reaction, or unit definition. Each relationship serves as a connection between two metabolic reactions, enabling easy exploration of different pathways and interactions between human metabolic data (Balaur et al., 2016). Using Neo4j with metabolic models can provide valuable insights into complex interactions, as it allows for the integration of heterogeneous data, such as chemical species, reactions, enzymes, and taxonomic data. This integration can aid in identifying key players across a wide range of biological application domains (Swainston et al., 2017).

More recently, based on the graph database MaSyMoS, which stores SBML and CellML models to represent biological systems in terms of functional, behavioral, and structural aspects (Henkel, Wolkenhauer & Waltemath, 2015), a protocol was developed to compare biochemical reaction networks (Lambusch et al., 2018). Nonetheless, analyzing its own SBML data using Neo4j remains a challenge. Biochem4j provides a web interface that does not allow users to analyze their data in their database. Recon2Neo4j and MaSyMoS created Java software using the JSBML library (Rodriguez et al., 2015) to import the data into their database. However, data loading is performed according to a predefined schema that retain several SBML components or properties that are targeted by their applications.

We are presenting neo4jsbml, a user-friendly Python package to import SBML files into a Neo4j database. Neo4jsbml utilizes tools developed by Neo4j to define the database schema, load data, query entities and relationships, and visualize the created graph. This package offers flexibility, extensibility, and ease of use in combining graph databases with SBML files. Flexibility is provided by leaving the choice of entities to be analyzed by the researcher. Extensibility is based on the intrinsic operation of neo4jsbml via introspection. Finally, neo4jsbml can be used through the command line and requires no programming skill. The graph can be enriched with heterogeneous data, allowing researchers to build, manipulate, annotate, and store their data with greater efficiency.

Material and Methods

Running method

The usage of the tool is described in Fig. 1.

It takes place in two stages. First of all, the user creates a schema defining which entities will be selected from the SBML model (Fig. 2). Building a schema requires certain rules, such as mapping the names of different items found in the schema to their corresponding names as defined in the SBML specifications. To illustrate this principle, node labels and node properties indicated in the schema need to match the name of the SBML component and the properties belonging to the component, respectively. Given this information, neo4jsbml identifies the objects and attributes to extract from the SBML, thanks to the library libsbml. Thus, only the entities required for analysis were loaded into Neo4j. SBML specifications indicate which components are linked to each other by storing an identifier corresponding to another component as an attribute. Therefore, neo4jsbml checks whether the two components are linked by following the strategies shown in Fig. 3. However, the Neo4j database operates on a directed graph; therefore, neo4jsbml infers directionality as indicated by the schema. Once the data are imported into Neo4j, the entities and relationships from the SBML model can be queried by Cypher through Neo4j.

Results

We developed neo4jsbml as a modular Python package that can be used as a standalone program or incorporated as a library in other programs. It is available through the Conda package management system (conda-forge community, 2015) and runs on all platforms without requiring extra privileges. SBML files are loaded by neo4jsbml, using the libsbml library (Bornstein et al., 2008), and the selected data will be loaded into the Neo4j database using the Python Neo4j driver, based on a database schema. Twelve packages are part or will be part of the SBML standard, but only eight have been fully implemented by the library libsbml. Neo4jsbml can consider four plugins: Flux Balance Constraints (Olivier & Bergmann, 2018), Groups (Hucka & Smith, 2016), Layout (Gauges et al., 2015) and Qualitative Models (Chaouiya et al., 2015) (Table S1). Importing data into Neo4j is done through pure Cypher queries, with no additional Neo4j plugins necessary.

To demonstrate the usefulness of neo4jsbml, we conducted a proof of concept study with three use cases.

Comparing two models

We compared two versions of the E. coli genome-scale metabolic model to highlight improvements between an older and a newer version from two repositories: BiGG and MetaNetX. Firstly, we compared the number of nodes and relationships in both models (Table 1). More than four thousand entities and twenty thousand relationships were loaded into Neo4j from the E. coli genome-scale models, iML1515 and iAF1260. Given one version of E. coli, the number of chemical species and reactions differed between the BiGG and MetaNetX databases. In addition, the iML1515 genome-scale metabolic model contained more nodes and relationships compared to the iAF1260 genome-scale metabolic model.

Table 1:

Number of entities and relationships loaded in Neo4j for the iAF1260 and iML1515 GEMs provided by the BiGG and MetaNetX databases.

Nodes are in lowercase, relationships are in uppercase.

	BiGG		MetaNetX
Node/Relationship	iAF1260	iML1515	iAF1260	iML1515
Compartment	3	3	4	4
Species	1668	1877	1976	2217
Parameter	8	5	7	5
Reaction	2382	2712	2374	2704
Model	1	1	1	1
UnitDefinition	1	1	1	1
GeneProduct	1261	1516	1262	1517
IN_COMPARTMENT	1668	1877	1976	2217
HAS_PRODUCT	4714	5328	4920	5529
HAS_LOWERFLUXBOUND	2382	2712	2374	2704
HAS_UPPERFLUXBOUND	2382	2712	2374	2704
HAS_PARAMETER	8	5	7	5
HAS_UNIT	8	5	7	5
IS_IMPLIED	3747	4624	3750	4627
IS_REACTANT	4517	5247	4553	5242

DOI: 10.7717/peerj.16726/table-1

Next, we analyzed the reactions associated with the fumarate metabolite (Fig. 4). The Cypher query is described in Eq. (1).

One additional reaction was linked to the fumarate metabolite in the iML1515 model compared to the iAF1260 model.

Enumerate metabolic pathways

The enumeration of metabolic pathways identifies and lists all the possible routes of biochemical reactions within a metabolic network. From the core model of E. coli, pathways involving the Formate and the Acetaldehyde metabolites were searched. Species, SpeciesReference and Reaction entities were loaded into Neo4j from the genome-scale model.

Then, graph reconstruction was performed by merging the Species and SpeciesReference nodes Eq. (4) and removing Species that were linked to fewer than four reactions Eq. (5) . Next, the neo4jefmtool plugin was used to enumerate the pathways involving the Fumarate and the Acetaldehyde metabolites Eq. (6). The two pathways identified are shown in the Fig. 6.

Discussions

Comparing genome-scale models, viewing metabolic networks and enumerating pathways are common applications in systems biology. On the one hand, biological processes are stored in the SBML format. On the other hand, Neo4j is a powerful graph database suited to link related entities, using the Cypher language. Neo4jsbml aims to conciliate both. The first use case showed the capability to perform quantitative and qualitative analysis on genome-scale models. Neo4j has a comprehensive set of features, as described by King et al. (2015) which include the ability to navigate and search through visualizations, design and customize pathway maps, and represent diverse data types. The second example highlights the ability of Neo4j to serve as an alternative viewer for visualizing metabolic pathways. The last illustration emphasizes the capacity of neo4jsbml to retain the required data and the strength of Neo4j to conduct specialized analyses.

Neo4jsbml employs a novel approach based on the creation of schema entities to identify entities that are useful for a dedicated application. Introspection is a programmatic principle that maps the schema of certain entities to a model to extract desired data. Therefore, neo4jsbml can deal with Level 2 Version 5 of the SBML standard (Fig. S3) and partially with SBML packages (Fig. S2). Our aim was to develop a tool that is as simple as possible, requiring only command-line skills.

However, as we observed for the use cases, the more complex the application, the more intricate the Cypher queries to perform graph refactoring or to extract the data. Additionally, programming skills may be required to develop extensions to perform dedicated analyses. In rare circumstances, data imported into Neo4j can produce unexpected results. The first scenario is shown in Fig. S3J. The model has two Event entities, each with one Event Assignment entity that supports the variable property. Without setting an id property for each Event Assignment, in the model, it would be difficult to infer the wishes of the user. The second scenario concerns the model presented in Fig. S2C. The Layout entity is linked to its Dimensions entity as well as to all other Dimensions entities found in the model. In these cases, a prior modification of the model or graph refactoring in Neo4j is required to meet expectations. In addition, if several models are loaded sequentially into the database, some entities can be the same, and the relationships between the models are mixed. Neo4jsbml alleviates this difficulty by adding an additional identifier to all imported entities.

The SBML standard represents mathematical expressions using MathML and annotations using the Resource Description Framework by adding XML elements to the model. Neo4jsbml extracts mathematical expressions as strings whereas annotations are loaded with XML tags.

However, neo4jsbml is highly dependent on the web application Arrows to create the schema of entities. To ensure a long-term future, neo4jsbml will need to support at least one other tool to provide the schema of entities. In addition, neo4jsbml uses simple Cypher queries to import data into Neo4j. It would be valuable to load data into another graph database using Cypher.

More broadly, introspection-based programming patterns can be used to extract information from other standards and import the required data into sql, no-sql or graph-oriented databases.

Conclusions

Neo4jsbml is a user-friendly Python package and it takes advantage of existing software to define the database schema, parse SBML files, and import the data into the Neo4j database. The three use cases demonstrated the benefits of visualizing SBML data as a graph. The biological processes, specified thanks to SBML standard, contain intricate interactions between biological entities, and Neo4j’s graph represents these relationships with nodes and edges, making it easier to analyze and visualize complex biological networks. Using Cypher, exploring data through statistics or creating subgraphs was intuitive. With the help of the schema database and introspection concept, only relevant information is loaded into the database. We foresee neo4jsbml as an essential tool for bridging the gap between SBML data and Neo4j. It will enable researchers to visualize and analyze relationships between different entities and fully leverage the capabilities of Neo4j.

Supplemental Information