Leveraging Horizontal Density Differences between Ontologies to Identify Missing Child Concepts: A Proof of Concept

Area of Science:

• Medical Informatics
• Ontology Engineering
• Bioinformatics

Background:

• Ontologies are crucial for organizing biomedical knowledge.
• Identifying concept discrepancies between ontologies aids knowledge integration.
• Previous work explored 'diamonds of concepts' with differing 'is-a' path lengths.

Purpose of the Study:

• To investigate if identical concepts in different ontologies share the same direct children.
• To quantify differences in child sets between SNOMED CT, NCIt, and other ontologies.
• To evaluate the feasibility of importing potentially missing child concepts algorithmically.

Main Methods:

• Comparative analysis of child concept sets for identical concepts across multiple ontologies.
• Quantitative data collection on child set discrepancies.
• Randomized controlled trials (RCTs) with human expert evaluation of 'fit for import' for candidate concepts.

Main Results:

• Significant differences were found in the child sets of identical concepts between SNOMED CT and NCIt relative to other ontologies.
• Quantitative data highlights the extent of these discrepancies.
• Two out of four RCTs demonstrated statistically significant support for the algorithmic import of missing child concepts.

Conclusions:

• Identical concepts can have divergent child sets across major biomedical ontologies.
• Algorithmic import of missing concepts is a viable strategy, supported by expert evaluation.
• Further research can refine methods for automated ontology alignment and knowledge enrichment.