AgroLD APIAgroLD API. Une architecture orientée services pour l'extraction de connaissances dans la base de données liées AgroLD

Gildas Tagny Ngompé, Aravind Venkatesan, Nordine El Hassouni, Manuel Ruiz, Pierre LarmandeIngénierie des Systèmes d Inf 2016 · DOI

TL;DR

This paper introduces AgroLD API, a service-oriented architecture designed to integrate and query diverse agronomic knowledge bases. By leveraging Semantic Web technologies and a RESTful API, AgroLD aims to bridge the gap in fragmented agricultural data, enabling bioinformaticians and domain experts to access and analyze complex biological information more effectively.

What Problem Does This Paper Address?

Agronomy is a complex, multidisciplinary field that generates vast amounts of data from various research areas like genetics, plant biology, and earth sciences. While high-throughput technologies have revolutionized data generation, this information often remains siloed in disparate, disconnected databases with varied formats. This fragmentation hinders efficient research, making it challenging to gain a comprehensive understanding of biological processes and their implications for crop improvement, disease resistance, and environmental impact. The paper addresses the need for a unified, accessible platform to integrate and query this dispersed agronomic knowledge.

Methods

The core of the AgroLD API is a service-oriented architecture (SOA) built upon Semantic Web technologies, specifically RDF and SPARQL. The API is designed using a RESTful approach, offering advantages in simplicity and interoperability with standard web technologies. Key components include:

  • Server Module: This acts as a mediator, implementing web services that query the AgroLD SPARQL endpoint and external services to fulfill requests. It's largely composed of services tailored for biological questions and was implemented using the Jersey library for RESTful Java development.
  • Interactive Documentation: A Swagger-generated interface provides a comprehensive, interactive description of each service, detailing its role, parameters, SPARQL query, return format, and HTTP status codes. This allows developers to easily understand and test the API.
  • JavaScript Client Library: A client library (swagger-client.js) enables JavaScript applications to interact with the API by treating services as methods of an object, abstracting away the complexities of direct URL calls and HTTP methods.
  • Dynamic Search Form: This feature allows users to build queries through a form-based interface, offering an iterative and exploratory search experience. It leverages the API to retrieve information and discover new research avenues based on previous results.
  • Integration of Existing Systems: The architecture integrates various existing tools and paradigms for semantic search, including faceted search (OpenLink Faceted), visual query builders (ReVeaLD), and natural language interfaces, aiming to provide a user-friendly experience for both biologists and bioinformaticians.

Datasets

The paper does not explicitly mention specific datasets used for training or evaluation in the context of machine learning models. However, the AgroLD project itself is designed to integrate data from various publicly available plant-centric data sources. The paper refers to the "AgroLD knowledge base" as the primary data source being queried and enriched.

Key Insights and Ideas

  • Bridging the Agronomic Data Divide: The paper highlights the critical need for integrating fragmented agronomic data, a challenge that has persisted despite advancements in data generation technologies.
  • SOA for Agronomic Knowledge: Proposing a service-oriented architecture for agronomic knowledge management is a novel approach, offering flexibility and extensibility.
  • RESTful API as a Standard: The adoption of a RESTful API standardizes access to complex biological data, making it more accessible to a wider range of applications and users.
  • User-Centric Design: The emphasis on providing user-friendly interfaces, such as dynamic search forms and interactive documentation, is crucial for enabling domain experts and bioinformaticians to effectively utilize the API.
  • Iterative and Exploratory Search: The dynamic search form promotes an iterative research process, allowing users to discover new information and refine their queries based on initial results.

Key Results

The paper focuses on the architectural design and implementation of the AgroLD API rather than presenting extensive quantitative results from large-scale experiments. However, it demonstrates the functionality of the implemented prototype through several use case scenarios:

  • Scenario 1 (Quick Search): A simple keyword search for "plant height" returned results in approximately 452 milliseconds, comparable to classical search engines.
  • Scenario 2 (Advanced Search): A more complex query involving an ontological identifier ("EO:0007403") and accessing associated QTLs took longer, with the form-based search requiring around 18 seconds (split into two stages).
  • Scenario 3 (External Service Query): A query requiring external service interrogation (e.g., finding publications for a protein) took approximately 24 seconds (split into two stages).
  • Scenario 4 (Relation Discovery): The system allows for the visualization of relationships between genes and pathways in a graph format, enhancing comprehension.

The results indicate that while simple searches are fast, more complex queries involving multiple steps or external service calls take longer. The presentation of results is generally in tabular format with hyperlinked entities for further exploration.

Limitations and Future Work

The authors acknowledge several limitations and outline future directions:

  • Development Time Constraints: The full development of a visual query system based on ReVeaLD, tailored to AgroLD's specific language, was not completed due to time constraints.
  • Language Specificity: The visual query system's language needs to be carefully defined to be user-friendly for biological concepts.
  • Performance Optimization: While the API demonstrates functionality, further performance improvements are planned.
  • Expanding Functionality: Future work includes improving performance and extending the API's functionalities to cover more entities represented in AgroLD, such as genomic annotations and homology information.
  • Text-Based Search: The authors plan to explore text-based indexing and search tools like ElasticSearch and SolR to enhance rapid search capabilities and hide technical details.
  • Integration of ReVeaLD: The integration of a ReVeaLD-based system with a specific language for AgroLD data is a key future development.

Summary updated

BibTeX
@article{ngompe2016agrold,
  author = {Gildas Tagny Ngompé and Aravind Venkatesan and Nordine El Hassouni and Manuel Ruiz and Pierre Larmande},
  title = {{AgroLD APIAgroLD API. Une architecture orientée services pour l'extraction de connaissances dans la base de données liées AgroLD}},
  journal = {Ingénierie des Systèmes d Inf.},
  year = {2016},
  doi = {10.3166/isi.21.5-6.133-157},
}
Figure