Prototype Semantic Infrastructure for Automated Small Molecule Classification and Annotation in Lipidomics

Leonid L. Chepelev, Alexandre Riazanov, Alexandre Kouznetsov, Hong Sang Low, Michel Dumontier, Christopher J. O. BakerBMC Bioinformatics 2011 · DOI

TL;DR

This paper introduces a prototype semantic web framework for automatically classifying and annotating small molecules, specifically lipids. By leveraging ontologies and semantic web services, it aims to overcome the manual curation bottleneck in lipidomics, enabling more efficient and accurate data analysis.

What Problem Does This Paper Address?

The explosion of high-throughput experimentation in lipidomics has led to an overwhelming amount of data on lipids and their derivatives. However, classifying and annotating these molecules remains a labor-intensive process, largely reliant on human curators. This manual approach is slow, prone to inconsistencies, and struggles to keep pace with the influx of novel, unclassified compounds. Accurate classification is crucial because a molecule's structure is intimately linked to its biological function, which in turn informs disease understanding and drug discovery. Existing classification schemes often lack formal, machine-readable definitions, making automated analysis difficult.

Methods

The core of the proposed solution is a semantic web infrastructure built upon two main components:

  • A Formal Lipid Ontology: This ontology extends existing lipid knowledge (like LiPrO and LIPID MAPS) with detailed, machine-readable definitions of lipid classes. These definitions are based on the presence and combination of specific functional groups within a molecule.
  • Federated Semantic Web Services: These services, deployed within the Semantic Annotation, Discovery, and Integration (SADI) framework, automate the classification process.
  • Structural Annotation Service: This service takes a chemical structure (represented as a SMILES string) and identifies and enumerates its functional groups.
  • Classifier Service: This service uses the functional group annotations and the formal lipid ontology to infer the appropriate lipid class for the molecule. It leverages OWL reasoning to check class descriptions against the identified functional groups.

The framework also integrates additional SADI services for retrieving associations between lipids and proteins, and for finding relevant literature, demonstrating a more comprehensive approach to lipid data integration. The SHARE client is used to orchestrate these services, allowing users to formulate queries without direct programming.

Datasets

  • LIPID MAPS Database: Used as a primary source for lipid class definitions and for evaluating the performance of the classification framework. A subset of 150 eicosanoid lipids from LIPID MAPS was used for evaluation.
  • LiPrO and Lipid Eicosanoid Ontology (LEO): These ontologies form the backbone of the classification system, providing formal definitions for lipid classes. LEO was specifically developed to address the eicosanoid hierarchy.

Key Insights and Ideas

  • Ontology-Driven Classification: The paper strongly advocates for the use of formal ontologies (specifically OWL) to define chemical classes. This shift from manual curation to logical, machine-readable definitions is key to enabling automated classification.
  • Functional Group as a Classification Basis: The framework demonstrates that a molecule's functional groups, when systematically identified and reasoned over, are powerful indicators for its classification.
  • SADI for Seamless Integration: The SADI framework proves instrumental in integrating disparate computational and informational resources. Its ability to automatically discover, compose, and execute services without programming significantly lowers the barrier to creating complex bioinformatics workflows.
  • Identifying Database Errors: The automated classification process, by comparing its results against expert-curated databases like LIPID MAPS, can highlight potential inconsistencies or errors in the existing data, leading to improved data quality.

Key Results

  • High Classification Accuracy: The prototype achieved a 96% classification accuracy for eicosanoids when compared to LIPID MAPS. This includes cases where the classification was to a more general, but still correct, class in the hierarchy.
  • Identification of Misclassifications: The framework successfully identified a case where a lipid was misclassified in the LIPID MAPS database, demonstrating its utility in data validation and refinement.
  • Efficient Workflow Orchestration: The SHARE client, in conjunction with SADI services, allowed for the execution of complex queries and workflows in under a minute, with service calls themselves taking less than a second.

Limitations and Future Work

  • Query Composition Complexity: While SADI simplifies service integration, composing SPARQL queries can still be challenging for users without programming experience. The authors suggest further work on user-friendly interfaces for query formulation.
  • Functional Classification: The current framework primarily relies on structural features. The authors acknowledge the need to extend this to handle functional classifications (e.g., toxicity, mutagenicity) by incorporating additional molecular annotations and corresponding services.
  • Ontology Maintenance: The functional group annotation service needs to be updated if new substructures become essential for defining lipid classes in the ontology. However, the SADI framework allows for the integration of new services to address this.
  • Scalability: While the current approach is efficient for the demonstrated use cases, the authors note that SPARQL querying over distributed services can face performance challenges similar to data federation.
  • Future Expansion: The authors envision expanding the framework to cover the entire LIPID MAPS hierarchy and to integrate more diverse services for predicting lipid metabolism, clustering similar chemicals, and even docking lipids to proteins.

Summary updated

BibTeX
@article{chepelev2011prototype,
  author = {Leonid L. Chepelev and Alexandre Riazanov and Alexandre Kouznetsov and Hong Sang Low and Michel Dumontier and Christopher J. O. Baker},
  title = {{Prototype Semantic Infrastructure for Automated Small Molecule Classification and Annotation in Lipidomics}},
  journal = {BMC Bioinformatics},
  year = {2011},
  doi = {10.1186/1471-2105-12-303},
}

Loading related content...

Figure