DOME Registry: implementing community-wide recommendations for reporting supervised machine learning in biology.

Area of Science:

• Life Sciences
• Computational Biology
• Bioinformatics

Background:

• Supervised machine learning (ML) is increasingly prevalent in biological research.
• Existing ML research often lacks standardized validation and transparent reporting, hindering reproducibility.
• The Data Optimization Model Evaluation (DOME) initiative provides recommendations for enhancing ML study rigor.

Purpose of the Study:

• To introduce the DOME registry, a centralized database for managing and accessing DOME-related information for published ML studies.
• To facilitate transparent and reproducible reporting of ML methods in the life sciences.
• To promote standardized evaluation of ML approaches through unique identifiers and DOME scores.

Main Methods:

• Development of the DOME registry (registry.dome-ml.org) as a database for ML study documentation.
• Integration with external resources (ORCID, APICURON, Data Stewardship Wizard) for streamlined annotation.
• Assignment of unique identifiers and DOME scores to publications within the registry.

Main Results:

• Establishment of a functional DOME registry for comprehensive documentation of ML studies.
• Demonstration of streamlined annotation processes using integrated external resources.
• Implementation of a DOME scoring system for standardized evaluation of ML publications.

Conclusions:

• The DOME registry provides a valuable resource for improving the transparency and reproducibility of ML in the life sciences.
• Community curation and adoption of DOME standards by publishers are crucial for future growth and impact.
• Continued refinement of DOME score definitions will further enhance the standardization of ML method evaluation.