Skip to content

Latest commit

 

History

History
51 lines (38 loc) · 3.18 KB

detailed-description.md

File metadata and controls

51 lines (38 loc) · 3.18 KB
Raw

Addendum: in-depth description of the automated trait mapping pipeline

Open Targets context

Data which we submit to Open Targets represents associations between traits and variants. In the case of ClinVar, traits are almost always inherited diseases. In the Open Targets evidence strings, they must be represented using terms from the Experimental Factor Ontology. EFO is an ontology (basically a controlled hierarchical dictionary) developed at EBI to standartise nomenclature of diseases and other terms.

Mapping free-text trait names to ontology terms is the only part of the Open Targets submission process which cannot be automated, and this is the reason it is contained as a separate protocol.

The idea is to run this protocol periodically and independently of the main submission protocol. While at submission time Open Targets data is always synchronised to an older, fixed ClinVar release, all new ClinVar data will be eventually incorporated into future Open Targets releases. Hence, it makes sense to do the curation work ahead of time and to decouple this process from the main evidence string generation.

To convert free-text descriptions to EFO terms, a trait mapping pipeline is used. You can find a diagram of the whole workflow here. The detailed description is presented below.

Querying ZOOMA

ZOOMA is first queried using the trait name. ZOOMA is a EBI-developed resource to take free-text trait descriptions as input and provide ontology terms as output. For example, if you query “Alzheimer's disease”, it should return http://www.ebi.ac.uk/efo/EFO_0000249. The pipeline uses ZOOMA results in the following manner:

  • If there are any high confidence mappings from EFO then they are output for use and no further processing is taken for that trait name.
  • If there are lower confidence mappings from EFO then these are output in a separate curation file, and the process stops here for that trait name.
  • If there are high confidence mappings not from EFO then their ontology IDs are used as input for querying OxO (see below)

Querying OxO

OxO is a tool which attempts to find ontology cross references for provided identifiers. It is used by the pipeline as follows:

  • Any EFO mappings found within a distance of 1 step are output for use, no further processing for this trait name.
  • Any EFO mappings within a distance greater than 1 step are output for curation.

Output

The output of the trait mapping pipeline consists of 2 files.

Automated trait mappings

The file automated_trait_mappings.tsv contains the following columns:

  • Trait name from ClinVar
  • Ontology URI
  • Ontology term label

Manual curation required

The file traits_requiring_curation.tsv contains the entries that require manual curation, including any trait name with no mappings. It contains the following columns:

  • Trait name from ClinVar
  • Frequency of this trait in ClinVar
  • ZOOMA mappings:
    • URI
    • Ontology label
    • Confidence
    • Datasource/annotator
  • OxO mappings:
    • URI
    • Ontology label
    • Confidence
    • Distance