GPT-4 Annotated Severity Scores for Human Phenotype Ontology Abnormalities

GPT-4 annotated the clinical severity of over 17,500 phenotypic abnormalities in the Human Phenotype Ontology across nine clinical characteristics. The annotations were benchmarked against ground-truth labels with a mean true positive recall rate of 97%. This dataset, created by Kitty B. Murphy and last updated in May 2026, provides quantitative severity metrics for prioritizing therapeutic targets in rare diseases.

Use Cases

• Ranking phenotypic abnormalities by health impact based on the integrated severity scoring system.
• Benchmarking LLM performance on clinical annotation tasks using the provided ground-truth comparisons.
• Prioritizing gene therapy targets for rare diseases based on automated severity metrics.
• Extending the annotation framework to incorporate additional clinical dimensions mentioned in the discussion.

Strengths

• Annotations cover over 17,500 phenotypic abnormalities across more than 8,600 rare diseases.
• Benchmarking showed strong performance with true positive recall rates ranging from 89% to 100%.
• Severity is operationalized using nine specific clinical characteristics and four frequency levels.
• Dataset is openly licensed under CC-BY-4.0.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The 1.8 KB file size suggests a very limited scope, likely containing summary scores rather than raw annotations.

Provenance

Source

figshare, authored by Kitty B. Murphy.

Collection Method

GPT-4 was employed to annotate phenotypic severity, with outputs benchmarked against ground-truth labels embedded in the HPO.

Freshness

Last updated 2026-05-21 05:44:10; freshness should be verified.