Enhancements to UniProtKB ingestion

[AlanSimmons]

Request:

Add the "Function" field to the data downloaded from UniProtKB.

[AlanSimmons]

Filtering to SwissProt curation only

Currently, UBKG only contains HGNC->UNIPROTKB mappings for proteins for which curation has been reviewed (i.e., SwissProt). TrEMBL curations are not ingested.

[AlanSimmons]

@computationdoc

Challenge: where to put function

Statement of problem

The current UNIPROTKB ETL maps as follows:

UniProt field	UBKG entity:property
Entry	Code node with CodeID=UNIPROTKB:Entry
Entry Name	Term node with name=Entry Name and relationship PT
Protein Names	Definition node with DEF=Protein Names, linked to the Code's associated Concept node.

The function is actually a type of definition; however, the Definition node is already being used for the Protein Names field.

Options:

1. Make the function a synonym

The UNIPROTKB ETL currently does not assign synonyms--i.e., Term nodes with relationship SYN to the code, based on the values in the node_synonyms field in the node_metadata file. The script could export the function field to the node_synonyms field.

• Pro: would not require changes to generation script code, and only a minor change to the existing ETL script.
• Con: This would overload the synonym role in that UNIPROTKB would have a special form of synonym. Mitigating this are the facts that the are currently no other synonyms for UNIPROTKB and the function text from UNIPROTKB starts with 'FUNCTION: '.

2. Build a term of different semantic type--e.g., FUNCTION

• Pro: This would be unambiguous.
• Con: This would require enhancing the generation framework to allow it to add terms of types other than PT/PT_SAB and SYN. This might entail regression testing of all ingestions for a use case that is specific to UNIPROTKB.

3. Assert new relationships between UNIPROTKB nodes and "function" nodes

• Con: Would require building a Code for each definition and then asserting a new relationship.

4. Make the function another Definition

• Pro: This aligns with the use of the Definition. The fact that the function text starts with 'FUNCTION:' would be a way to distinguish the function from the current definition.
• Con: The generation script assumes only one Definition per code. This option would require enhancing the generation script to allow for multiple definition nodes, similar to how it handles multiple synonym nodes.

Recommendation

Option 1. Currently, only UNIPROTKB codes have both definitions and functions in UBKG.

[AlanSimmons]

Result of assigning function to synonym

Example. The function is the term of type SY

[AlanSimmons]

New decision: assign function to definition; short name to synonym

The image is an annotated screen capture of a UniProtKB detail page, describing the changes to make in UBKG.

[AlanSimmons]

Synonym arrangement

UniProtKB stores names and synonyms for proteins in the Protein Names field of the downloaded file. The field is delimited with parentheses--e.g., Approved Name (1st synonym)(2nd synonym) etc.

The script:

• uses the approved name for the Term of type PT
• uses all other names for Term nodes of type SY

In addition, the script uses the value of the Entry Name field from the download as a synonym.

[AlanSimmons]

Result

Example showing the approved name, all synonyms, and the definition.

[AlanSimmons]

Dependency on https://github.com/x-atlas-consortia/hs-ontology-api/issues/14