TL;DR
The Cell Ontology (CL) is a crucial resource for standardizing cell type definitions, enabling FAIR data principles in the rapidly growing field of single-cell omics. This paper details CL's current applications, ongoing improvements, and future directions, highlighting its role in integrating diverse data modalities and advancing our understanding of cellular diversity.
What Problem Does This Paper Address?
The explosion of single-cell omics data (transcriptomics, proteomics, etc.) has created an unprecedented opportunity to understand cellular diversity. However, the sheer volume and heterogeneity of this data pose significant challenges for integration, analysis, and hypothesis generation. Biologists, clinicians, and bioinformaticians need a standardized way to annotate and connect this data to existing biological knowledge. This paper addresses the need for a robust, species-agnostic framework for cell type annotation that can bridge the gap between classical biological knowledge and the new insights from single-cell omics. The Cell Ontology (CL) has emerged as a pivotal resource for achieving FAIR (Findable, Accessible, Interoperable, and Reusable) data principles by providing these standardized terms.
Methods
The paper describes how the Cell Ontology (CL) is developed and utilized. Key methods include:
- Ontology Development: CL is built using standardized methods, adhering to OBO Foundry principles, and integrates with other ontologies like the Gene Ontology (GO) and Uberon (anatomy).
- Term Definition: CL primarily defines classical cell types based on structural, functional, and molecular criteria. It is evolving to incorporate "T-types" defined by transcriptomic profiles, using evidence criteria like anatomical location, morphology, functional properties, and multimodal evidence from multiple reference datasets.
- Data Integration: CL serves as a backbone for integrating data from various platforms and consortia (e.g., HuBMAP, HCA, CELLxGENE). It facilitates interoperability through mappings to species-specific ontologies.
- Automated Annotation: CL is increasingly integrated into automated cell type annotation tools (e.g., Azimuth, CellTypist, OnClass, Sfaira), leveraging its structure and relationships to improve classification accuracy.
- LLM Integration: Large Language Models (LLMs) are being used to assist in generating and refining cell type definitions, summarizing literature, and identifying candidate links to GO terms. Tools like CLARA (CL Accuracy Reference Agent) are developed to validate LLM-generated content.
- Taxon Constraints: CL formally represents taxonomic constraints to ensure assertions about cell types are valid across species, preventing incorrect assertions.
- Cross-Species Bridging: CL bridges species-specific ontologies (e.g., Drosophila Anatomy Ontology) using methods like SSSOM to create unified hierarchies and facilitate cross-species comparisons.
Datasets
The paper discusses the use of various datasets and resources that rely on or contribute to the Cell Ontology:
- Single-cell RNA-sequencing (scRNAseq) data: This is a primary data type driving the development and use of CL, particularly from large atlasing efforts like the Human Cell Atlas (HCA) and the Brain Initiative Cell Atlas Network (BICAN).
- CELLxGENE Corpus: A large, consistently annotated database of single-cell transcriptomics data that mandates the use of CL IDs for annotation.
- HuBMAP Human Reference Atlas: Provides expert input and curated data for CL development.
- Reference Datasets: CL incorporates T-types based on evidence from integrated sc/snRNAseq atlases from multiple sources, with community agreement on reference datasets.
- Literature Data: LLMs are used to extract information from scientific literature to support definition generation and validation.
No new benchmarks are explicitly introduced, but the paper highlights how CL enables the use and integration of existing large-scale datasets.
Key Insights and Ideas
- CL as a FAIR Data Enabler: CL is fundamental to making single-cell omics data Findable, Accessible, Interoperable, and Reusable by providing standardized identifiers and definitions.
- Bridging Classical and Transcriptomic Definitions: CL is actively working to harmonize classical cell type definitions with those derived from transcriptomic data (T-types), recognizing the value of both approaches.
- The Power of Ontological Relationships: Beyond simple definitions, the relationships within CL (e.g., part_of, capable_of) create a knowledge graph that supports complex queries, gene enrichment analysis, and machine learning applications.
- LLMs as Curation Assistants: LLMs are transforming ontology development by accelerating definition generation and literature review, though expert human review remains critical.
- Importance of Community Collaboration: The development and maintenance of CL are highly collaborative, involving numerous institutions, researchers, and subject matter experts.
- Cross-Species Harmonization is Crucial: CL's ability to bridge species-specific ontologies is essential for understanding conserved biological principles and comparative genomics.
Key Results
- Widespread Adoption: CL is mandated by major consortia and platforms like HuBMAP, HCA, and CELLxGENE for cell type annotation, enabling better data integration and findability.
- Content Expansion: CL has significantly expanded its content, adding over 1000 new cell type terms and revising definitions to reflect evolving domain knowledge.
- Improved Quality Control: Technological advancements and tools like the Ontology Development Kit (ODK) and automated checks have led to more efficient editing and higher quality releases.
- LLM Integration Success: LLMs have been successfully used to generate draft definitions and assist in curation, with tools like CLARA providing an additional layer of validation.
- Cross-Species Integration: CL supports integration across diverse species, illustrated by its use in the EBI Single Cell Expression Atlas for cross-species data integration.
Limitations and Future Work
- Defining T-types is Challenging: Establishing consensus on reference datasets and analysis for defining T-types is difficult due to experimental and computational differences across studies.
- Naming T-types: Reconciling diverse names from different datasets and ensuring names reflect community expectations is an ongoing challenge.
- Taxon Constraints: Ensuring CL does not assert anything about a cell type that is not valid across all species where it exists requires careful formalization of taxonomic constraints.
- Future Work:
- Continue to incorporate T-types and transcriptomic hierarchies.
- Further develop LLM-assisted workflows for more efficient and accurate ontology development.
- Enhance CL's role in bridging data modalities (e.g., imaging, connectomics) and across species.
- Improve the scalability and automation of CL development to keep pace with the rapid growth of single-cell omics data.
- Expand CL's coverage and quality to remain relevant for future biological research.
Summary updated