Abstract
Taxonomy has long struggled with analyzing vast amounts of phenotypic data due to computational and accessibility challenges. Ontology-based technologies provide a framework for modeling semantic phenotypes that are understandable by computers and compliant with FAIR principles. In this paper, we explore the use of Phenoscript, an emerging language designed for creating semantic phenotypes, to produce computable species descriptions. Our case study centers on the application of this approach to dung beetles (Coleoptera: Scarabaeinae). We illustrate the effectiveness of Phenoscript for creating semantic phenotypes. We also demonstrate the ability of the Phenospy python package to automatically translate Phenoscript descriptions into natural language (NL), which eliminates the need for writing traditional NL descriptions. We introduce a computational pipeline that streamlines the generation of semantic descriptions and their conversion to NL. To demonstrate the power of the semantic approach, we apply simple semantic queries to the generated phenotypic descriptions. This paper addresses the current challenges in crafting semantic species descriptions and outlines the path towards future improvements. Furthermore, we discuss the promising integration of semantic phenotypes and nanopublications, as emerging methods for sharing scientific information. Overall, our study highlights the pivotal role of ontology-based technologies in modernizing taxonomy and aligning it with the evolving landscape of big data analysis and FAIR principles.
Introduction: The "Dark Data" Problem
For centuries, taxonomists have described new species using Natural Language (NL)—essentially, paragraphs of descriptive text. While effective for human communication, this format is a "dark data" trap for computers. If a researcher wants to compare the leg shapes of hundreds of different beetle species, they currently must manually read and extract data from hundreds of papers. This lack of interoperability violates the FAIR principles (Findable, Accessible, Interoperable, Reusable).
The challenge lies in the complexity of biological morphology. Translating the subtle curve of a beetle’s thorax into a rigid machine-readable format without losing nuance is a massive technical hurdle. This research addresses this gap by providing a bridge between human-readable descriptions and machine-actionable logic.
Methodology: The Phenoscript and Phenospy Pipeline
The authors introduce a computational pipeline centered on Phenoscript, a newly designed domain-specific language that allows taxonomists to "code" a species description. The workflow consists of several key stages:
- Semantic Coding: Instead of writing a paragraph, the researcher uses a VS Code plugin to write Phenoscript statements (e.g.,
organism > head > color >> red). - The Phenospy Pipeline: A Python package called Phenospy converts these statements into the Web Ontology Language (OWL), creating a "knowledge graph" (an ABox) of the specific specimen.
- Ontology Integration: The descriptions are mapped to standardized biological ontologies, specifically the Insect Anatomy Ontology (AISM) and the Coleoptera Anatomy Ontology (COLAO).
- Automated Translation: Crucially, the pipeline can translate these "code" descriptions back into natural language for traditional publication. This ensures the taxonomist only has to write the description once, with the NL version acting as a byproduct of the structured data.
- Micro-CT Imaging: To ensure high-fidelity data, the team utilized micro-CT scanning to observe internal and external structures in 3D, providing the empirical basis for the semantic models.
Case Study: Grebennikovius Dung Beetles
The study focuses on a proof-of-concept dataset involving four species of the dung beetle genus Grebennikovius (Scarabaeinae) from Tanzania. To accommodate the specific morphology of these beetles, the authors introduced 152 new terms into the AISM and COLAO ontologies. The final semantic descriptions consist of 756 ontology individuals, representing discrete phenotypic statements.
Key Insights: Individual-Based Modeling and Anatomical Rigor
A core innovation of this method is the individual-based (ABox) approach. Unlike previous "class-based" models that described species in the abstract, this method describes a specific specimen. This is more intuitive for taxonomists and better for tracking biological variation.
Furthermore, the use of ontologies enforces anatomical rigor. Traditional terms like "outer margin" of a leg are often anatomically imprecise; ontologies require using "dorsal" or "posterior" based on true evolutionary positions. This transition ensures that taxonomic data is "born-digital" and fully FAIR-compliant.
Results: Queryable Taxonomy and Nanopublications
The team demonstrated that they could use SPARQL (a query language for databases) to instantly extract data across species. For example, they could instantly count that G. armiger has 32 distinct shape-related traits compared to 25 in G. basilewskyi. The automated translation from OWL to Natural Language was largely successful, producing descriptions that resemble traditional taxonomic text while remaining hyperlinked to ontology definitions.
The paper also explores the concept of nanopublications—tiny, discrete units of publishable information (e.g., "Species X lives in Forest Y") that are immediately queryable by the global scientific community. This represents a shift toward a more granular and integrated scientific record.
Limitations and Future Outlook
While promising, the workflow faces several challenges:
- Learning Curve: Writing in Phenoscript is currently slower than writing in plain English, creating a barrier to entry for taxonomists unfamiliar with coding environments.
- Ontology Gaps: The need to manually add 152 terms highlights that biological ontologies are not yet "saturated."
- Intraspecific Variation: The current model is optimized for single specimens; modeling variation within a species (e.g., "usually red, but sometimes brown") requires more complex data models.
- Complex Logic: Automated translation still struggles with complex relative comparisons (e.g., "wing shorter than in species X").
Conclusion
Ontology-based technologies are pivotal in modernizing taxonomy. By moving from unstructured text to structured knowledge graphs, the field can finally integrate with the evolving landscape of big data analysis, ensuring that the "dark data" of the past becomes the accessible knowledge of the future.
Summary updated