biopragmatics / semra Goto Github PK

we got burned by the HP mapping from its specific neoplasms (e.g., neoplasm of the rectum; https://hpo.jax.org/app/browse/term/HP:0100743) to NCIT's high-level neoplasm term (NCIT:C3262), so some paths go through HPO's neoplasm terms (that are specific) then to NCIT, then back to some other HPO term, which means that you can basically call any neoplasms equivalent.

here's an example showing this

hp:0010788      skos:exactMatch doid:1984       semapv:MappingChaining  doid,hp,umls                    1.0             doid:1984 umls:C0034885 mesh:D012004 hp:0100743 ncit:C3262 hp:0010788
hp:0011750      skos:exactMatch doid:1984       semapv:MappingChaining  doid,hp,umls                    1.0             doid:1984 umls:C0034885 mesh:D012004 hp:0100743 ncit:C3262 hp:0011750
hp:0011752      skos:exactMatch doid:1984       semapv:MappingChaining  doid,hp,umls                    1.0             hp:0011752 ncit:C3262 hp:0100743 umls:C0034885 ncit:C3350 doid:1984
hp:0012289      skos:exactMatch doid:1984       semapv:MappingChaining  doid,hp,umls                    1.0             hp:0012289 ncit:C3262 hp:0100743 umls:C0034885 ncit:C3350 doid:1984
hp:0012720      skos:exactMatch doid:1984       semapv:MappingChaining  doid,hp,umls                    1.0             doid:1984 umls:C0034885 mesh:D012004 hp:0100743 ncit:C3262 hp:0012720
hp:0012777      skos:exactMatch doid:1984       semapv:MappingChaining  doid,hp,umls                    1.0             hp:0012777 ncit:C3262 hp:0100743 umls:C0034885 ncit:C3350 doid:1984

1. Need to make sure we don't incorporate many to many relations
   • technically the priority mapping only forces the subjects to be unique, but there should be some intermediate where we say that for each object, there can only be one subject from each namespace that maps to it.
2. Need to filter out evidence paths that have two conepts from the same namespace

• #4
• #3
• Consider adding default header with some information like mapping_tool (semra) and an updated description.
• Add unit test to semra that runs sssom.validate() on a constructed mapping set

From biopragmatics/pyobo#128:

https://icd.who.int/browse11/Downloads/Download?fileName=mapping.zip

why don't we start a repo of scripts that grab random resources and convert them into SSSOM? PyOBO is already doing this, but it could have less baggage associted

see https://mapping-commons.github.io/sssom/Mapping/

Given a set of mappings, they can be stratified on three axes:

1. mapping justification (semapv:ManualMappingCuration and maybe semapv:UnspecifiedMatching vs. others like semapv:LexicalMatching, semapv:BackgroundKnowledgeBasedMatching, semapv:MappingInversion, and semapv:MappingChaining)
2. mapping set
3. source-target prefix pair

Some ideas on making a comparison in order of increasing complexity

1. Compare mapping overlap to get a simple precision, recall, and $F_1$
2. Penalize incorrect 1-many, many-1, and many-to-many mappings
3. Incorporate ontology hierarchy

semapv:MappingChaining
semapv:UnspecifiedMatching