TL;DR
This paper introduces ORNASEQ, a new ontology designed to systematically capture the complex data provenance of next-generation RNA sequencing (RNAseq) experiments. By leveraging existing ontologies and providing a real-world dataset, ORNASEQ aims to improve the reproducibility and understanding of RNAseq studies, which are becoming increasingly vital in research and diagnostics.
What Problem Does This Paper Address?
Next-generation RNA sequencing (RNAseq) is a powerful technology that generates vast amounts of data, crucial for understanding biological processes and disease. However, the complexity of RNAseq experiments, involving numerous steps, reagents, software versions, and researchers, makes it incredibly challenging to track and document all the details that contribute to the final data. This "data provenance" – the history and origin of data – is essential for ensuring data quality, reproducibility, and for diagnosing issues when results are unexpected. Without a standardized way to capture this provenance, it's difficult to manage, query, and reuse RNAseq data effectively, especially as the scale of experiments grows from dozens to thousands or even tens of thousands of samples. This lack of a robust provenance system hinders scientific progress and can be a barrier to clinical applications where regulatory compliance is paramount.
Methods
The core of this work is the development of the Ontology for RNA Sequencing (ORNASEQ). This is an application ontology, meaning it's tailored for a specific domain – in this case, RNAseq data provenance. ORNASEQ was built by leveraging and extending existing ontologies, primarily the Ontology for Biomedical Investigations (OBI) and other ontologies from the Open Biological and Biomedical Ontology (OBO) Foundry. This approach ensures that ORNASEQ is grounded in established semantic frameworks, promoting interoperability.
The authors describe ORNASEQ as containing 162 terms in total. A significant portion (117 terms) are derived from 16 existing ontologies, demonstrating a strategy of reuse and integration. Additionally, 28 new terms were added to OBI, and 17 terms were defined directly within ORNASEQ. This hybrid approach allows ORNASEQ to be both comprehensive for RNAseq and to contribute back to broader biomedical ontologies. The ontology is designed to annotate RNA-based next-generation sequencing, but its principles are also applicable to DNA-based sequencing.
Datasets
The paper introduces a significant real-life sample provenance dataset. This dataset comprises:
- 1,347 next-generation sequencing samples.
- 93 data fields, utilizing ORNASEQ terms.
- Data processed using four different versions of a primary analysis pipeline, with each pipeline version having one to three variants (sub-versions).
- Samples processed as either "single-end" or "paired-end".
- Alignment performed using either STAR or Bowtie, leading to nine possible pipeline variants.
This dataset is provided in two formats:
- PROV-XML (Additional file 2), a standard for representing provenance information.
- An Excel table (Additional file 1) summarizing the data.
Crucially, the dataset intentionally includes incorrect and missing data to reflect real-world scenarios and to serve as a benchmark for developing robust provenance tools that can handle imperfect data.
Key Insights and Ideas
- The critical need for standardized provenance in RNAseq: As RNAseq experiments scale up and become more integrated into diagnostics, the ability to track every step of data generation and analysis is no longer a luxury but a necessity for reliability and trust.
- Leveraging existing ontologies is key to interoperability: Instead of building from scratch, ORNASEQ builds upon established biomedical ontologies, ensuring its terms can be understood and integrated with other biological data resources.
- Provenance capture is challenging in both wet lab and dry lab: While informatics steps are machine-readable, the complexity of software versions, parameters, and log files still presents significant challenges. Wet lab provenance, often documented in notebooks or disparate files, is even more difficult to standardize.
- A hybrid approach to provenance capture is practical: The paper proposes capturing provenance by storing complete protocol definitions as files (e.g., PDFs) alongside structured, hardcoded fields for critical or frequently changing parameters. This balances detail with queryability.
- Real-world data is messy: The inclusion of intentionally incorrect and missing data in their sample dataset highlights that any practical provenance system must be robust enough to handle imperfections.
Key Results
- Development of ORNASEQ: A new application ontology specifically designed for RNA sequencing data provenance, containing 162 terms derived from existing ontologies and new definitions.
- Creation of a comprehensive sample provenance dataset: A real-world dataset of 1,347 RNAseq samples, meticulously documented with 93 data fields using ORNASEQ terms, and representing various pipeline versions and configurations.
- Demonstration of provenance complexity: The dataset illustrates the intricate variations in experimental and informatics processes, underscoring the need for sophisticated provenance tracking and workflow management systems.
- Provision of a benchmark dataset: The PROV-XML and Excel versions of the dataset serve as a valuable resource for the community to develop and test new provenance tools and data models.
Limitations and Future Work
- Scope of ORNASEQ: While ORNASEQ is designed for RNA-based sequencing, the authors note that much of it would also apply to DNA-based next-generation sequencing, suggesting potential for broader application or further development.
- Complexity of RNAseq studies: The paper acknowledges that RNAseq studies are complex and will continue to evolve, with hundreds of application-specific protocol variants emerging (e.g., ATAC-seq, Drop-Seq). This implies that ORNASEQ, while a strong foundation, may need continuous expansion.
- Need for workflow management systems: The authors emphasize the need for potent workflow management systems and smart clustering algorithms to help segment incomplete and erroneous data, indicating that ORNASEQ is a foundational component for such future systems.
- Future development of provenance models and tools: The provided dataset is intended to be useful in designing future workflow systems and provenance models, suggesting that this work is a starting point for further research and development in the field.
Summary updated