Hierarchy-Aware Hyperbolic and Semantic Reranking for Ontology-Based Phenotype Linking

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Science & Research — Life Sciences & Biology, Health & Medical Research · Depth: Expert, quick

Summary

A new hierarchy-aware workflow is introduced for ontology-based phenotype linking, addressing the challenge of extracting structured knowledge from clinical narratives, particularly for concepts within complex hierarchical ontologies like the Human Phenotype Ontology. This approach integrates Large Language Models for span detection with retrieval and a hybrid reranking strategy. The reranking combines Euclidean (semantic) and hyperbolic (hierarchical) embeddings, both trained on HPO. While hyperbolic embeddings alone do not surpass standard semantic retrieval, they offer complementary structural signals. When combined with Euclidean representations, the hybrid method significantly improves performance over strong baselines, outperforming existing methods and producing more hierarchically coherent predictions, especially for implicit phenotype mentions. Experiments on the public ID-68 benchmark and a new CHU-50 clinical dataset, both publicly released, demonstrate performance gains and better alignment with ontology structure. The authors also introduce a hierarchy-aware evaluation framework.

Key takeaway

For Machine Learning Engineers developing clinical text analysis systems, you should consider integrating hybrid reranking strategies for phenotype linking. Your current semantic retrieval methods can be significantly enhanced by combining Euclidean and hyperbolic embeddings, especially when dealing with complex hierarchical ontologies like HPO. This approach yields more accurate and hierarchically coherent predictions, particularly for subtle or implicit phenotype mentions, improving diagnostic precision. Explore the publicly released CHU-50 dataset and hierarchy-aware evaluation framework to validate your models.

Key insights

A hybrid reranking approach using semantic and hyperbolic embeddings enhances phenotype linking in hierarchical ontologies, particularly for implicit mentions.

Principles

• Hyperbolic embeddings provide structural signals.
• Hybrid embeddings enhance hierarchical coherence.
• Ontology structure improves prediction quality.

Method

The workflow uses LLMs for span detection, followed by retrieval and a hybrid reranking strategy. This strategy combines Euclidean (semantic) and hyperbolic (hierarchical) embeddings trained on the Human Phenotype Ontology.

In practice

• Improve diagnostic precision in genomics.
• Better handle implicit phenotype mentions.
• Use hierarchy-aware evaluation metrics.

Topics

• Phenotype Linking
• Clinical Text Analysis
• Large Language Models
• Hyperbolic Embeddings
• Human Phenotype Ontology
• Knowledge Extraction

Best for: NLP Engineer, AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• HypRAG: Hyperbolic Dense Retrieval for Retrieval Augmented Generation — Hyperbolic embeddings, with geometry-aware pooling, significantly improve RAG by preserving language's intrinsic hierarchical structure.
• From Annotated Clinical Narratives to Ontology: Structuring Brazilian Portuguese Clinical Data — A new ontology structures Brazilian Portuguese clinical narratives, enhancing NLP interoperability and semantic expressiveness.
• BeLink: Biomedical Entity Linking Meets Generative Re-Ranking — Instruction-tuning generative models significantly enhances Biomedical Entity Linking re-ranking efficiency and accuracy.
• Systematically decoding pathological morphologies and molecular profiles with unified multimodal embedding — Multi-Embed unifies morphological and molecular data for superior disease understanding and prediction.
• HypOProto: Hyperbolic Ordinal Prototypes for Left Ventricular Filling Pressure Classification — Hyperbolic geometry enhances interpretability and ordinal encoding for medical image classification.
• Modeling semantic association in self-paced reading with language model embeddings — Methodological choices, especially embedding model type, critically impact semantic association quantification in reading comprehension.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.