Introduction
Topological Data Analysis (TDA) is a powerful but highly interdisciplinary field, with applications scattered across biology, medicine, materials science, and finance. For researchers, finding specific technical precedents—such as the use of "Persistent Homology" on "EEG time series"—is often difficult using standard search engines like Google Scholar or PubMed, which lack the necessary technical granularity.
To address this, the authors developed DONUT (Database of Original & Non-Theoretical Uses of Topology), a specialized, searchable repository designed to catalog real-world applications of TDA. DONUT provides a structured "map" of the field, helping practitioners find precedents, assisting grant writers in justifying TDA's utility, and organizing the community's collective knowledge.
Methodology and Infrastructure
The authors implemented a multi-layered pipeline to move beyond simple spreadsheets and create a robust research tool:
- Backend Management: The system utilizes Zotero as the primary bibliographic database, facilitating collaborative entry and metadata management.
- The "Tags & Flavors" System: Every entry is manually classified into three categories: Area of Application (e.g., medicine), Mathematical Tools (e.g., Mapper, Clustering), and Input Type (e.g., point cloud). Additionally, papers are labeled with "flavors": Confirm (where TDA matches existing methods) and Innovate (where TDA provides novel results or handles unique data).
- Search Engine: The platform uses Xapian, an open-source, language-aware search engine. This allows the system to handle spelling mistakes (e.g., "simplical" vs. "simplicial") and transliterated names effectively.
- Frontend Interface: A web interface built with Flask allows users to query the database using refined operators like
tag:orauthor:, providing a sub-millisecond search experience.
Database Composition and Insights
The DONUT database has grown from a modest list of 10 papers in 2017 to over 431 cataloged entries. The data is largely crowdsourced from the TDA community, including significant contributions from experts like Prof. Mikael Vejdemo-Johansson.
Analysis of the database reveals several key insights into the state of the field:
- The "Innovate" Bar: Only 58 out of 431 entries were labeled with the "innovate" flavor. The authors maintain strict criteria for this tag to ensure it remains a meaningful indicator of TDA's unique value proposition.
- Tool Prevalence: The database highlights the dominance of specific tools within the community, with Persistent Homology being the most frequently cataloged method.
- Taxonomy as a Service: In the absence of a formal bibliographic format for TDA, DONUT serves as an evolving, dynamic taxonomy for the entire field.
Key Results and Future Directions
The project has successfully launched donut.topology.rocks, offering automated normalization, spelling suggestions, and international character support. An experimental "landscape visualization" feature also helps users find similar papers through a visual metaphor rather than text alone.
Despite its success, the project faces challenges. The manual curation of "Areas of Application" is subjective and time-consuming. Future work aims to implement a hierarchical "tree visualization" for tags and full-text indexing of open-access PDFs. As the volume of TDA literature continues to expand, the authors are looking toward community-driven sustainability to maintain the high quality of the database's manual tagging.
Summary updated