Skip to content

Welcome to Phentrieve

Phentrieve is a comprehensive system for mapping clinical text in multiple languages to Human Phenotype Ontology (HPO) terms via a Retrieval-Augmented Generation (RAG) approach. The system supports multilingual text processing, benchmarking across various embedding models, and provides flexible interfaces through a Python package, API, and web frontend.

Key Features

  • Multilingual HPO Term Mapping: Map clinical text to HPO terms in multiple languages without translation
  • Advanced Text Processing: Process clinical text with semantic chunking and assertion detection
  • Multiple Embedding Models: Support for domain-specific, language-specific, and general multilingual models
  • Cross-Encoder Re-Ranking: Improve retrieval precision with specialized re-ranking models
  • Comprehensive Benchmarking: Evaluate and compare model performance with detailed metrics
  • Multiple Interfaces: Command-line tools, FastAPI backend, and Vue.js frontend

Core Concept

In clinical genomics and rare disease diagnosis, identifying phenotypic abnormalities in patient descriptions is a critical step. Traditional approaches often require translation when descriptions are in languages other than English, which can introduce inaccuracies.

Phentrieve implements a novel approach using multilingual embedding models that map semantically similar concepts from different languages to nearby points in the embedding space. This allows direct matching between non-English clinical descriptions and English-based HPO terminology.

Dive Deeper