TL;DR
The BioHackathon 2015 brought together a diverse group of researchers to tackle the challenges of integrating, reusing, and reproducing life sciences data. This paper summarizes their efforts in developing semantic technologies and tools to make biological data more FAIR (Findable, Accessible, Interoperable, and Reusable), ultimately aiming to accelerate scientific discovery.
What Problem Does This Paper Address?
The life sciences generate an overwhelming amount of complex data, from genomic sequences to proteomic assays and clinical phenotypes. However, this data is often siloed, uses inconsistent formats and terminologies, and lacks clear provenance, making it difficult to integrate, analyze, and reproduce research findings. This fragmentation hinders scientific progress and the development of truly interoperable bioinformatics tools and knowledge bases. The BioHackathon 2015 aimed to address these issues by fostering collaboration and developing semantic web technologies to create a more unified and accessible data landscape.
Methods
The paper details numerous projects undertaken at the BioHackathon, each employing various semantic web technologies and approaches. Key methods include:
- Ontology Development and Application: Creating and extending ontologies (e.g., ORTH for orthology, PIERO for chemical transformations, GlycoRDF for glycomics) to provide standardized vocabularies and semantic meaning to biological data.
- RDF Representation and Linked Open Data (LOD): Converting diverse data formats (e.g., VCF files, UniProt data, proteomic assay annotations) into RDF triples to enable integration and querying through SPARQL.
- Semantic Web Services and APIs: Developing services that expose biological data and tools through standardized interfaces, facilitating programmatic access and integration.
- Natural Language Processing (NLP) and Text Mining: Applying NLP techniques to extract phenotypic information from clinical text and drug labels, and developing tools like "Crick-chan" for natural language querying of biomedical literature.
- Workflow Management and Reproducibility Tools: Working on technologies like the Common Workflow Language (CWL) and Docker to ensure that computational analyses are reproducible and can be shared and executed across different platforms.
- Knowledge Graph Annotation: Developing tools like ORKA to enable human curation and annotation of knowledge graphs, improving data quality and provenance.
- Data Integration and Harmonization: Strategies for linking and harmonizing data from disparate sources, such as chemical databases and disease ontologies.
Datasets
The paper describes work on various datasets, often involving the conversion or annotation of existing resources. Notable examples include:
- 1000 Genomes Project Phase 3 Variant Call Format (VCF) files: Used for building genomic variation graphs.
- GRCh37 human reference genome: Used in conjunction with VCF files.
- UniProt database: Used for evaluating OWL representation of proteins.
- InParanoid, OMA, and TreeFam datasets: Used for orthology ontology development.
- FossilCalibrations database: Used for molecular evolutionary process calibration.
- DisGeNET and EBI data: Integrated for disease-phenotype association analysis.
- PubMed abstracts: Used for text mining and the "Crick-chan" system.
- Drug labels: Used for extracting drug effects and indications.
- Arabidopsis thaliana mutant data: Used for plant metabolome database development.
No new benchmarks were explicitly introduced, but the work focused on making existing data more accessible and reusable.
Key Insights and Ideas
- The Power of Ontologies for Interoperability: The consistent theme is that well-defined ontologies are crucial for giving semantic meaning to biological data, enabling its integration and reuse across different systems.
- Semantic Web Technologies are Maturing for Life Sciences: The hackathon demonstrated that technologies like RDF, SPARQL, and OWL are not just theoretical but are being actively applied to solve real-world problems in bioinformatics.
- Reproducibility is a Multi-faceted Challenge: Beyond just data, ensuring the reproducibility of computational analyses requires robust tools for workflow management, software versioning, and environment standardization (e.g., CWL, Docker, Guix).
- Human Curation Remains Vital, but Needs Semantic Support: Tools like ORKA highlight the ongoing need for human expertise in annotating and validating biological data, and how semantic technologies can enhance this process by capturing provenance and enabling reuse of annotations.
- Bridging the Gap Between Text and Structured Data: NLP and text mining are essential for unlocking the vast knowledge embedded in unstructured biomedical literature, making it computable and linkable to structured databases.
- The "FAIR" Principles are Gaining Traction: The paper emphasizes the growing importance of making data Findable, Accessible, Interoperable, and Reusable, and how the projects contributed to tools and practices that support these principles.
Key Results
- Genomic Variation Graphs: Modeled genomic variation graphs using RDF semantics, demonstrating efficient loading and querying.
- Phenotype Ontology Translation: Translated over 11,000 HPO terms into Japanese to aid in EHR analysis.
- Orthology Ontology (ORTH): Developed an ontology to capture orthology concepts and converted existing datasets into RDF.
- Proteomics Data Representation: Explored OWL 2 EL for representing protein semantics, demonstrating its scalability for large datasets.
- Metabolomics Tools: Developed web applications and tools for visualizing and analyzing metabolomics data.
- NLP for Phenotype Extraction: Developed pipelines for extracting phenotype associations from journal articles and drug labels.
- Reproducibility Tools: Extended CWL support and worked on Docker container registries and GNU Guix for reproducible software deployment.
- Semantic Haiku Generation: Created a prototype system for generating haikus from semantic web data.
Limitations and Future Work
- Scalability of OWL: While OWL 2 EL was explored, certain complex features of OWL were not expressible, and further analysis is needed on how this approach scales to very large ontologies.
- Ambiguity in Data Formats: The FossilCalibrations database, while useful, lacked explicit semantics, leading to ambiguity in data interpretation.
- Integration Challenges: The paper notes that while many tools and ontologies were developed, further work is needed to fully integrate them and address issues like mapping inconsistencies and the need for standardized metadata.
- NESCent Closure: The paper incorrectly states NESCent was active in 2020, when it had shut down in 2015, highlighting a need for careful temporal context in reporting.
- Future Work: The authors suggest continued development of ontologies, semantic web services, NLP tools, and reproducibility frameworks. Specific future directions include evaluating the developed ontologies, expanding data integration efforts, and developing quantitative measures for FAIRness.
Summary updated