Abstract
In digital chemistry, the lack of standardized metadata for chemical assertions creates a significant trust gap for researchers utilizing public APIs. This paper introduces a provenance model designed to capture the origin, authorship, and methodology of scientific claims. Through a systematic analysis of 10.8 million Nanopublications, 150,000 Wikidata records, and 14 Minimal Information Standards (MISTS), the authors identified 23 essential metadata properties across five key dimensions: Authorship, Temporal aspects, Source, Generating process, and Evidence. This model provides a roadmap for API developers to improve data integrity and machine-readability in chemical research.
Introduction
Computational workflows in chemistry increasingly rely on data retrieved from public APIs. However, these assertions—such as a specific melting point for a compound—often lack the necessary context to evaluate their quality. Without knowing the experimental method, the author, or the date of the claim, researchers cannot automatically assess the integrity of integrated findings. Currently, there is no universal standard for reporting this "provenance," leading to inconsistent terminology and a lack of clear guidelines for data producers.
The Proposed Model
To address this, the authors developed a conceptual framework that categorizes provenance into five major dimensions: Authorship, Source, Generating Process, Evidence, and Temporal aspects. This structure ensures that both the human and machine agents involved in a claim are properly credited and contextualized.
Methods
The study employed a three-pronged approach to define a robust metadata model:
- Nanopublications Analysis: The authors queried 10.8 million structured knowledge snippets to identify real-world metadata usage patterns.
- Wikidata Analysis: Approximately 150,000 "Chemical Compound" records were extracted to analyze how crowdsourced platforms handle references and claims.
- MISTS Review: 14 "Minimal Information Standards" from FAIRsharing.org were analyzed to determine the biomedical community's core reporting requirements.
The identified properties were pruned and mapped to standard ontologies like PROV-O (Provenance Ontology) and PAV (Provenance, Authoring and Versioning) to ensure machine-readability across platforms.
Results and Key Insights
The analysis revealed a significant disparity in how different platforms handle metadata. Nanopublications generally provide more comprehensive information regarding evidence and authorship compared to Wikidata. However, a critical "black hole" was identified across both platforms: the Generating Process.
As shown in the comparison, while "Source" information is prevalent, details regarding the specific experiments or analytical procedures (Generating Process) are almost entirely absent in current datasets. This highlights a major gap in current data reporting practices that the proposed model seeks to rectify.
The final model consists of 90 unique properties, with a core subset of 23 designated as "essential." These include assertedBy (Author), assertedOn (Date), retrievedFrom (Database), and supportingAnalysisMethod.
Conclusion
This preliminary model, implemented in JSON-LD and RDF, offers a standardized path forward for chemical data provenance. By providing a format-agnostic framework, it allows API developers to prioritize essential metadata. Future work involves stress-testing the model within the BioThings API (specifically MyChem.info) using live data from prominent chemical sources. Long-term success will depend on the broader biomedical community adopting these standards to ensure the transparency and reproducibility of digital research.
Summary updated