Supervised segmentation of phenotype descriptions for the human skeletal phenome using hybrid methods

Area of Science:

• Bioinformatics
• Computational Biology
• Medical Informatics

Background:

• Ontology-based formalization of phenotype descriptions is crucial for capturing complex biological knowledge.
• Phenotype descriptions inherently combine anatomical entities and qualities, requiring segmentation into atomic elements.
• Existing methods necessitate improved approaches for accurate and efficient phenotype data processing.

Purpose of the Study:

• To develop and evaluate a novel two-phase hybrid segmentation method for phenotype descriptions.
• To segment phenotype descriptions into their atomic elements, specifically anatomical entities and qualities.
• To assess the impact of external resources on segmentation performance.

Main Methods:

• A two-phase hybrid segmentation approach combining individual classifiers.
• Utilized aggregation schemes including set operations and simple majority voting.
• Tested on a corpus of skeletal phenotype descriptions from the Human Phenotype Ontology.

Main Results:

• The best hybrid method achieved an F-Score of 97.05% in phase I (segmentation of anatomical entities and qualities).
• Phase II achieved F-Scores of 97.16% and 94.50% for different aggregation schemes.
• Segmentation performance was not significantly affected by the inclusion or exclusion of domain dictionaries.

Conclusions:

• The developed hybrid segmentation method demonstrates high accuracy in processing phenotype descriptions.
• External resources like domain dictionaries do not critically impact the initial segmentation phase.
• The effectiveness of hybrid methods is contingent on specific data characteristics and the defined goals.