METRIN-KG: A knowledge graph integrating plant metabolites, traits, and biotic interactions

This paper introduces METRIN-KG, a knowledge graph that integrates heterogeneous data on plant metabolites, traits, and biotic interactions to facilitate efficient data retrieval and support research in life sciences.

The Gist

Imagine the natural world as a massive, bustling city. In this city, every plant is a unique building, every insect is a resident, and the chemicals they produce are the tools and materials they use to survive.

For a long time, scientists trying to understand this city had to visit three completely different libraries that didn't talk to each other:

1. The Chemistry Library: Stacks of books on the millions of tiny chemicals plants make (metabolites).
2. The Architecture Library: Blueprints of plant shapes, sizes, and features (traits like leaf size or seed weight).
3. The Social Network Library: Records of who eats whom, who pollinates whom, and who fights whom (biotic interactions).

The problem? A researcher looking for a specific plant might find its chemical recipe in one library, its height in another, and its enemy in a third. Connecting the dots required hours of manual detective work, and often, the clues were lost in translation.

Enter METRIN-KG: The "Super-App" for Nature.

This paper introduces METRIN-KG (MEtabolomes, TRaits, and INteractions-Knowledge Graph). Think of it not as a library, but as a massive, intelligent GPS map that finally connects all three libraries into one seamless network.

Here is how it works, using simple analogies:

1. The Great Translator (Ontologies)

Imagine trying to have a conversation between a chef, a carpenter, and a social worker. They all use different words for similar things.

• The chef says "flour."
• The carpenter says "sawdust."
• The social worker says "raw material."

In science, one database might call a plant part a "leaf," while another calls it "foliage," and a third uses a complex code. METRIN-KG acts as a universal translator. It uses a special "dictionary" (called an ontology) to ensure that when the system sees "leaf," "foliage," or a code, it knows they all mean the same thing. It cleans up the messy data so everything speaks the same language.

2. The Big Connector (The Knowledge Graph)

Once the data is translated, METRIN-KG builds a giant web.

• The Nodes (The Dots): These are the plants, the chemicals, and the insects.
• The Edges (The Lines): These are the connections.

The Magic: Before, you had to ask, "What chemicals does this plant make?" and then separately ask, "What bugs eat this plant?"
With METRIN-KG, you can ask a complex question like: "Show me all plants that are short, grow in dry soil, produce a specific chemical, and are eaten by a specific moth."

The system instantly draws a line from the plant to its chemical, then from the chemical to the moth, and from the moth back to the plant's height. It reveals patterns that were previously invisible.

3. Real-World Detective Work (Case Studies)

The authors tested this "Super-App" with real detective cases:

• Case 1: The Endangered Species Detective.
  They asked the system to find all "Near Threatened" plants and see what traits they have, what chemicals they produce, and who interacts with them. It's like a conservationist asking, "If we save this rare plant, what else in the ecosystem depends on it?" The system instantly pulled up a profile of 9,000+ species, showing which ones are at risk and what makes them unique.

• Case 2: The Drug Discovery Sleuth.
  Imagine a scientist looking for a new medicine. They know a certain chemical (Onopordopicrin) might fight cancer. They ask METRIN-KG: "Who makes this chemical, and who eats them?" The system found a specific plant (Eriophyllum confertiflorum) that produces it and revealed a complex web of insects interacting with it. This helps scientists understand why the plant makes the chemical (maybe to fight off those insects) and where to find it.

• Case 3: The Farmer's Assistant.
  Farmers use "push-pull" strategies (planting one crop to push pests away and another to pull them in). METRIN-KG helped map out exactly which plants produce chemicals that repel specific pests, helping to design better, chemical-free farming systems.

Why This Matters

Think of METRIN-KG as Google for the entire web of life, but instead of just finding webpages, it finds the hidden relationships between a plant's DNA, its physical shape, the chemicals it brews, and its neighbors.

• For Scientists: It saves years of work. Instead of reading 1,000 papers to connect the dots, they can run a query in seconds.
• For Policymakers: It helps make better decisions on conservation by showing how saving one plant might save an entire ecosystem.
• For Everyone: It brings us closer to understanding the "chemistry of life" and how nature solves problems, which could lead to new medicines, better crops, and a healthier planet.

In short, METRIN-KG takes the scattered puzzle pieces of the natural world and snaps them together to show us the full picture.

Technical Summary

1. Problem Statement

The paper addresses a critical fragmentation in biodiversity data management. While vast amounts of data exist regarding plant metabolomes (chemical compounds), traits (measurable characteristics like height or leaf area), and biotic interactions (relationships with other organisms), these datasets are currently:

• Isolated: Stored in separate, heterogeneous databases (e.g., TRY for traits, GloBI for interactions, ENPKG/LOTUS for metabolites).
• Disconnected: Lack semantic interoperability, making it difficult to query relationships across these domains (e.g., "Which traits are associated with specific defense metabolites in plants interacting with a specific herbivore?").
• High-Dimensional: The sheer volume and complexity of chemical diversity (estimated at 1.5–25.7 million plant metabolites) combined with ecological data pose significant analytical challenges.
• Format-Locked: Much data remains in unstructured formats (PDFs, Excel sheets) or pairwise correlation metrics that obscure mechanistic links.

The authors argue that understanding the "chemistry of life" is essential for linking species traits and interactions to ecosystem functioning, but current tools cannot effectively bridge these three pillars.

2. Methodology

The authors developed METRIN-KG (MEtabolomes, TRaits, and INteractions-Knowledge Graph), a semantic knowledge graph that integrates three major data sources using a rigorous data engineering pipeline:

A. Data Sources

1. Metabolomes: Enriched datasets from the Experimental Natural Products Knowledge Graph (ENPKG) (1,600 tropical plant extracts) and LOTUS (via Wikidata).
2. Traits: Functional trait data from the TRY database (41 specific traits including leaf area, seed mass, photosynthesis rates, etc.).
3. Interactions: Pairwise interaction data from Global Biotic Interactions (GloBI).

B. Data Processing & Harmonization

• Taxonomy Mapping: Scientific names from all sources were mapped to Wikidata identifiers (QIDs) to ensure a unified taxonomic backbone. This involved querying 15 different taxonomy databases used by GloBI.
• Metadata Normalization (Semantic Matching): A significant challenge was the unstructured nature of metadata in GloBI (e.g., "body part," "life stage," "biological sex" containing compound terms like "2 guts" or "10M,9F").
  • The authors developed a script using Python (Owlready2, SentenceTransformers) to parse these terms.
  • They utilized the all-MiniLM-L6-v2 model to generate dense vector embeddings for input terms and ontology terms.
  • Cosine similarity was calculated to match raw text to standardized ontology concepts (UBERON, Plant Ontology, PATO, etc.). Matches with a score $\ge$ 0.7 were flagged for manual curation.
• Unit Standardization: Trait units from TRY were mapped to the QUDT (Quantities, Units, Dimensions, and Types) vocabulary.

C. Ontology Design

• The authors utilized and extended the Earth Metabolome Initiative (EMI) Ontology.
• This ontology integrates concepts from SOSA (Sensor, Observation, Sample, Actuator), RO (Relation Ontology), and ENPKG.
• It defines new concepts (e.g., emi:NonTrait) to handle structured values and ensures semantic interoperability across chemical, biological, and ecological domains.

D. Knowledge Graph Construction

• Subgraph 1 (Metabolites): Built using Ontop, a virtual knowledge graph system. Tabular data from ENPKG was loaded into a relational database (MySQL), and mappings were defined between the relational schema and the EMI ontology to materialize RDF triples.
• Subgraphs 2 & 3 (Traits & Interactions): Built using Python rdflib. TSV files from TRY and GloBI were processed, linked via Wikidata identifiers, and converted directly into RDF triples based on the EMI schema.
• Indexing & Querying: The integrated graph was indexed using Qlever (a high-performance SPARQL engine). A user-friendly SPARQL editor (based on qlever-ui) was implemented, featuring fallback mechanisms to retrieve entity summaries even if external web resources are missing.

3. Key Contributions

1. First Integrated Resource: METRIN-KG is the first resource to semantically link plant metabolomes, functional traits, and biotic interactions in a single, queryable graph.
2. Semantic Interoperability: The successful application of NLP (sentence embeddings) to map unstructured metadata (body parts, life stages) to controlled ontologies, enabling complex cross-domain queries.
3. Accessibility Tools:
   • A public SPARQL endpoint for programmatic access.
   • Integration with ExpasyGPT, an LLM-driven tool that allows non-technical users to query the graph using natural language, which is then translated into SPARQL to avoid LLM hallucinations.
4. Open Science: All code, ontologies, RDF files, and processed datasets are publicly available via GitHub and Zenodo under open licenses (GPL-3.0, CC0).

4. Results & Case Studies

The authors demonstrated the utility of METRIN-KG through five case studies (CS) covering diverse research fields:

• CS1 (Conservation Science): Retrieved traits, interactions, and metabolites for 9,299 species with "Near Threatened" IUCN status.
  • Finding: While interaction data was available for 8,598 species, trait data was sparse (1,051 species), and metabolite data was very limited (47 species), highlighting data gaps for conservation planning.
• CS2 (Functional Ecology): Listed traits for plants producing diterpenoids.
  • Finding: Identified 2,131 species with 98 traits; Pinus sylvestris had the highest measurement frequency.
• CS3 (Human Health/Pharmacology): Mapped producers of onopordopicrin (a potential antimicrobial/cytotoxic compound) and their biotic interactions.
  • Finding: Identified Eriophyllum confertiflorum as a hub species with 77 incoming connections, revealing complex interaction networks.
• CS4 (Sustainable Agriculture): Investigated allelopathic interactions in push-pull agriculture.
  • Finding: The query successfully filtered out 45% of erroneous fungal interactions (misidentified as allelopaths) and identified 67 valid tripartite interactions involving crops, parasites, and protective plants.
• CS5 (Theoretical Ecology): Linked Leaf Economic Spectrum (LES) traits with metabolites.
  • Finding: Only 49 species had data for both LES traits and metabolites, underscoring the current scarcity of integrated data and the value of the graph in identifying these rare overlaps.

5. Significance

• Bridging Disciplines: METRIN-KG enables researchers to move beyond siloed studies, allowing for the formulation of hypotheses that connect chemical mechanisms (metabolites) to ecological outcomes (traits and interactions).
• Drug Discovery & Agriculture: By linking specific metabolites to ecological contexts (e.g., defense against specific herbivores), the graph aids in prioritizing plant species for drug discovery or sustainable agricultural practices (e.g., biocontrol).
• Data-Driven Ecology: It provides a framework for "Functional Traits 2.0," where metabolomics is treated as a core functional trait, essential for understanding ecosystem functioning under climate change.
• Scalability: The modular architecture allows for the future integration of additional data types (e.g., soil data, climate data, image data) and expansion beyond the Plantae kingdom.

In conclusion, METRIN-KG represents a significant step toward a holistic, FAIR (Findable, Accessible, Interoperable, Reusable) biodiversity knowledge infrastructure, transforming fragmented data into a connected network capable of answering complex life-science questions.