Embeddings Examples

Notebooks demonstrating the Embedding Provider Interface: ontology term embeddings from OLS, LLMs and the ontology closure, and how they compare with classic semantic similarity.

• Ontology Term Embeddings in OAK
  • Connecting to OLS
  • Vectors as numpy and pandas
  • Pairwise similarity
  • Nearest neighbours and text search
  • Classic semantic similarity as vectors: the closure model
  • Comparing the similarity structure of models
  • Comparing sets of terms
  • LLM embeddings computed on demand
  • Command line
  • See also
• Subsumption Recapitulation: Do LLM Embeddings Encode the Ontology Hierarchy?
  • Sampling term pairs
  • Vectors for every model
  • 1. Does embedding similarity track ontology similarity?
  • 2. Is a term’s parent closer than its sibling?
  • Where the models get it wrong
  • Summary
• Phenotype Profile Matching: Classic Semantic Similarity vs LLM Embeddings
  • Disease profiles and simulated patients
  • The OAK API, for one patient and one disease
  • Vectors and scores
  • Results
  • Discussion
• Where Ontology Similarity and Text-Embedding Similarity Disagree
  • Sampling pairs
  • Scoring
  • The scatter plots
  • The corners
  • What the corners show