Skip to contents

Retrieve pre-computed EDAM term embeddings from AnnotationHub

Source: R/embed.R
get_edam_embeddings.Rd

Downloads (and caches locally via AnnotationHub) a matrix of text-embedding-3-small embeddings for all non-deprecated EDAM terms. Each term is represented by its label concatenated with its oio:hasDefinition text. On first call the file is downloaded; subsequent calls in the same or future sessions use the local cache.

Usage

get_edam_embeddings()

Value

a list with components ids, labels, types, texts, embeddings (numeric matrix, terms × dimensions), model, and created.

Details

Lookup order:

  1. If EDAM_EMBEDDING_RDS is set to a readable .rds path, that file is loaded and returned immediately.

  2. The file bundled with the package at inst/demo_embedding/edam_embeddings.rds is used (via system.file).

  3. AnnotationHub is queried for a biocEDAM embedding resource.

To override the bundled demo file with a freshly generated artifact, set EDAM_EMBEDDING_RDS to its path or ensure the resource is in AnnotationHub (see make_edam_embeddings).

Examples

emb <- get_edam_embeddings()
#> Loading bundled EDAM embeddings from /private/var/folders/yw/gfhgh7k565v9w83x_k764wbc0000gp/T/RtmpbnJ5Gy/temp_libpathed7e7dd7ef17/biocEDAM/demo_embedding/edam_embeddings.rds
cat(sprintf("%d terms | %d dimensions | model: %s\n",
    length(emb$ids), ncol(emb$embeddings), emb$model))
#> 2399 terms | 1536 dimensions | model: text-embedding-3-small