Developing a Knowledge Graph for Pharmacokinetic Natural Product-Drug Interactions

Sanya B. Taneja, Tiffany J. Callahan, Mary F. Paine, Sandra L. Kane‐Gill, Halil Kilicoglu, Marcin P. Joachimiak, Richard D. Boyce

TL;DR

This paper introduces NP-KG, a novel knowledge graph that integrates biomedical ontologies and scientific literature to identify potential pharmacokinetic interactions between natural products and drugs. This framework aims to accelerate the discovery of these interactions, which are crucial for preventing adverse drug events, by providing a computational tool for researchers.

What Problem Does This Paper Address?

The increasing popularity of natural products, often used alongside pharmaceutical drugs, raises concerns about potential natural product-drug interactions (NPDIs). Specifically, pharmacokinetic NPDIs, where a natural product alters a drug's absorption, distribution, metabolism, or excretion, can lead to reduced efficacy or unexpected adverse effects. While significant research exists for drug-drug interactions, understanding NPDIs, particularly their underlying mechanisms, remains a challenge. Existing computational approaches for NPDIs are limited, highlighting the need for a more comprehensive framework to mine and integrate knowledge from diverse sources.

Methods

The authors developed a novel knowledge graph framework called NP-KG. This framework integrates:

  • Biomedical Ontologies: Standardized terminologies and relationships from the Open Biological and Biomedical Ontology (OBO) Foundry.
  • Publicly Available Databases: Including drug interaction databases (DIKB, Drug Central, FDA Drug Interaction database) and chemical entity databases (ChEBI).
  • Scientific Literature: Full texts of research articles related to green tea and kratom.

The construction involved:

  1. Data Integration: Using the Phenotype Knowledge Translator (PheKnowLator) workflow to semantically integrate ontologies and linked data. Extensions were created for natural product information.
  2. Semantic Relation Extraction: Employing SemRep and INDRA/REACH systems to extract subject-relation-object triples (predications) from the full text of scientific literature.
  3. Literature-based Graph Construction: Standardizing extracted relations to OBO Foundry ontologies, applying symmetric and transitive closure to infer new relationships, and creating a graph from these predications.
  4. KG Merging: Combining the ontology-grounded KG and the literature-based graph to form the final NP-KG.
  5. Evaluation: Using two strategies:
  6. Knowledge Recapturing: Comparing direct edges and shortest paths in NP-KG against a curated ground truth dataset (NaPDI Center) for known interactions.
  7. Meta-path Discovery: Applying path searches in NP-KG to generate hypotheses for specific NPDIs and comparing them to published clinical results.

Datasets

  • Biomedical Ontologies: OBO Foundry ontologies (e.g., Mondo Disease Ontology, Human Phenotype Ontology, Gene Ontology, ChEBI).
  • Drug Databases: Drug Interaction Knowledge Base (DIKB), Drug Central, FDA Drug Interaction database.
  • Natural Product Data: Extensions created by the authors based on Global Substance Registration System.
  • Scientific Literature: Full texts of PubMed-indexed articles related to green tea and kratom.
  • Ground Truth Data: Curated data from the NaPDI Center database, containing results from in vitro experiments and clinical pharmacokinetic NPDI studies.

Key Insights and Ideas

  • Bridging the Gap: NP-KG successfully bridges the gap between structured biomedical knowledge (ontologies, databases) and unstructured, rich information found in scientific literature, a critical step for understanding complex biological interactions.
  • Mechanistic Hypotheses Generation: The framework demonstrates the ability to generate plausible mechanistic hypotheses for NPDIs, moving beyond simple identification of interactions to understanding how they occur.
  • Contradictions as Opportunities: The study acknowledges that contradictions within the literature and extracted data are inevitable. Instead of discarding them, the authors highlight the need for manual review and future work to resolve these, potentially leading to deeper insights.
  • Human-Machine Collaboration: The authors envision NP-KG as a tool to enhance human-machine collaboration, guiding researchers and accelerating the discovery process for NPDIs.

Key Results

  • The integrated NP-KG contains 745,512 nodes and 7,249,576 edges.
  • Knowledge Recapturing:
    • Green Tea: 38.98% congruence, 15.25% contradiction with ground truth.
  • Kratom: 50% congruence, 21.43% contradiction with ground truth.
  • Meta-path Discovery: Identified potential pharmacokinetic mechanisms for several NPDIs, including green tea-raloxifene, green tea-nadolol, kratom-midazolam, kratom-quetiapine, and kratom-venlafaxine, which were congruent with published literature.
  • The framework successfully identified known interactions and provided insights into potential mechanisms for these interactions.

Limitations and Future Work

  • Limitations:
    • No manual filtering of search strategies, potentially leading to non-relevant predications.
  • Known errors in relation extraction systems (e.g., handling negation, identifying substrates).
  • Entity linking in the literature-based graph only mapped 60-70% of subjects and objects.
  • Meta-path searches did not incorporate information about metabolites or physicochemical properties.
  • Limited availability of ground truth data for NPDI validation.
  • Future Work:
    • Address limitations of relation extraction systems.
  • Extend NP-KG to include more natural products and knowledge domains.
  • Apply improved hypothesis generation and evaluation methods.
  • Incorporate contextual information (dosage, negative results, confidence scores) to resolve contradictions.
  • Expand to include other pharmacokinetic targets (e.g., nuclear receptors).
  • Develop more comprehensive validation strategies.

Summary updated

Abstract

NP-KG is the first KG to integrate biomedical ontologies with full texts of the scientific literature focused on natural products. We demonstrate the application of NP-KG to identify known pharmacokinetic interactions between natural products and pharmaceutical drugs mediated by drug metabolizing enzymes and transporters. Future work will incorporate context, contradiction analysis, and embedding-based methods to enrich NP-KG. NP-KG is publicly available at https://doi.org/10.5281/zenodo.6814507. The code for relation extraction, KG construction, and hypothesis generation is available at https://github.com/sanyabt/np-kg.
BibTeX
@article{taneja2023developing,
  author = {Sanya B. Taneja and Tiffany J. Callahan and Mary F. Paine and Sandra L. Kane‐Gill and Halil Kilicoglu and Marcin P. Joachimiak and Richard D. Boyce},
  title = {{Developing a Knowledge Graph for Pharmacokinetic Natural Product-Drug Interactions}},
  journal = {Journal of Biomedical Informatics},
  year = {2023},
  doi = {10.1016/j.jbi.2023.104341},
}

Loading related content...

Figure