Back

Human Phenotype Ontology (HPO) Mapper: Semantic Mapping of Clinical Findings to the Human Phenotype Ontology Using AI-Powered Embeddings and LLM-Based Quality Control

Kadhim, A. Z.; Green, Z.; Boags, A.; George, M.; Heinson, A.; Stammers, M.; Kipps, C. M.; Beattie, R. M.; Robinson, P. N.; Ashton, J. J.; Ennis, S.

2025-12-22 health informatics
10.64898/2025.12.20.25342726 medRxiv
Show abstract

O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=111 SRC="FIGDIR/small/25342726v1_ufig1.gif" ALT="Figure 1"> View larger version (36K): org.highwire.dtl.DTLVardef@8edb94org.highwire.dtl.DTLVardef@f20105org.highwire.dtl.DTLVardef@21033corg.highwire.dtl.DTLVardef@15b865e_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOVISUAL ABSTRACT:C_FLOATNO C_FIG Structured phenotypic annotations linked to genetic data can drive diagnostic insight and therapeutic discovery in complex diseases. However, poor research access to the rich clinical data trapped in unstructured clinical records remains a significant barrier to phenotype-genotype integration. Here, we present Human Phenotype Ontology (HPO) Mapper, a scalable AI-assisted tool designed to ingest semantically structured clinical findings paired with anatomical region and accurately map them to HPO terms and associated genes. We applied HPO Mapper to two forms of standardised clinical input extracted from inflammatory bowel disease (IBD) patient records. The first data type consisted of paired clinical findings + anatomical regions derived from unstructured clinical reports and the second was standardised ICD-10 code-derived phenotypes. HPO Mapper achieved high semantic alignment and mapping accuracy for both data types (F1 = 0.85 {+/-} 0.05 and 0.84 {+/-} 0.03, respectively). Our publicly available tool enables real-time HPO mapping for clinical applications providing a foundation for scalable AI-driven phenotyping across diseases.

Matching journals

The top 9 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.