Semantic Type Matching

[erikyao]

A CUI can have multiple semantic types. E.g.

CUI	concept_name	semantic_type
C3539881	gamma-interferon	phsu
C3539881	gamma-interferon	gngm
C3539881	gamma-interferon	aapp
C3539881	gamma-interferon	imft

When a retired CUI is mapped to a multi-semantic-typed CUI, it's reasonable to match the semantic types for precise replacement. E.g.

CUI	concept_name	semantic_type	➡️	CUI	concept_name	semantic_type	replacement
C0021740	Recombinant Interferon-gamma	phsu	➡️	C3539881	gamma-interferon	phsu	⭕
C0021740	Recombinant Interferon-gamma	gngm	➡️	C3539881	gamma-interferon	gngm	⭕
C0021740	Recombinant Interferon-gamma	aapp	➡️	C3539881	gamma-interferon	aapp	⭕
C0021740	Recombinant Interferon-gamma	imft	➡️	C3539881	gamma-interferon	imft	⭕
C0021740	Recombinant Interferon-gamma	phsu	➡️	C3539881	gamma-interferon	gngm	❌

However, Colleen also mentioned that some highly related semantic types should be considered as matched. E.g.

CUI	concept_name	semantic_type	➡️	CUI	concept_name	semantic_type	replacement
C1335188	PAG gene	gngm	➡️	C1705981	PAG1 wt Allele	gngm	⭕
C1335188	PAG gene	aapp	➡️	C1705981	PAG1 wt Allele	gngm	❓

The aapp ➡️ gngm match is also worth consideration.

Colleen and I came to the idea that:

1. If the new CUI is multiple-semantic-typed, enable semantic type matching. E.g. (C0021740,aapp) ➡️ (C3539881,aapp).
2. If the new CUI has only one semantic type, disable semantic type matching. E.g. (C1705981,aapp) ➡️ (C3539881,gngm)

@newgene @andrewsu do you have any idea on the matching conditions? Or shall we carry out exact matching for all replacement? How about explicitly whitelisting? Appreciate your thoughts!

Note that the replacement can occur to either subjects or objects, so the matching conditions may affect the semantic meaning of those involved predicates.

[erikyao]

Andrew's comment in Dec 20 Meeting:

• If the old semtype is found in the new CUI's semtypes, match.
• If not found, remove the whole predication.
• I.e. exact semtype matching all the time.

Yao's TODO: report no. of removed predications.

[erikyao]

Total number of predications: $90,703,432$. This number is obtained after:

1. Predictions with 0 novelty scores are removed
2. Predictions with invalid subject names are removed (just a few)
3. All piped CUIs are separated to individual predictions. E.g. C0056207|3075 will be counted as in two predictions.
4. Predictions with deleted CUIs (according to UMLS) are removed.

N.B. Node normalizer lookups not performed yet.

Now consider predictions with replaced CUIs:

• W/o exact semtype matching, $4,740,841$ prediction should be replaced.
• With exact semtype matching, $4,070,917$ prediction should be replaced.
• That means $669,924$ predications would be discarded, $0.74\%$ of the total no. of predications

[andrewsu]

For posterity (and only if it's easy to generate), can you provide some summary of the 669,924 predications that would be discarded? For example, are most of those because aapp got changed to gngm or vice versa?

But regardless, I'm comfortable moving forward with the exact semtype matching...

[erikyao]

Top 10 discarded subject semtypes

SUBJECT_SEMTYPE	SUBJECT_SEMTYPE_NAME	count
hcro	Health Care Related Organization	116904
fndg	Finding	27473
dsyn	Disease or Syndrome	24603
tisu	Tissue	23088
genf	Genetic Function	21068
lbpr	Laboratory Procedure	18223
patf	Pathologic Function	17526
gngm	Gene or Genome	16133
bpoc	Body Part, Organ, or Organ Component	10744
ortf	Organ or Tissue Function	10507

Top 10 discarded object semtypes

OBJECT_SEMTYPE	OBJECT_SEMTYPE_NAME	count
patf	Pathologic Function	37985
genf	Genetic Function	36366
dsyn	Disease or Syndrome	32800
ortf	Organ or Tissue Function	26308
fndg	Finding	26290
sosy	Sign or Symptom	17063
gngm	Gene or Genome	12314
aapp	Amino Acid, Peptide, or Protein	12219
anab	Anatomical Abnormality	11565
lbpr	Laboratory Procedure	7657

Top 10 discarded predicates

predicate	count
LOCATION_OF	221965
PROCESS_OF	83473
AFFECTS	61902
COEXISTS_WITH	37521
TREATS	33710
USES	28081
CAUSES	25475
PART_OF	23851
AUGMENTS	20763
ASSOCIATED_WITH	17609

Top 20 discarded predications (as in triples of (subject_semtype, predicate, object_semtype))

SUBJECT_SEMTYPE	SUBJECT_SEMTYPE_NAME	PREDICATE	OBJECT_SEMTYPE	OBJECT_SEMTYPE_NAME	count
hcro	Health Care Related Organization	LOCATION_OF	resa	Research Activity	53532
hcro	Health Care Related Organization	LOCATION_OF	lbpr	Laboratory Procedure	31750
hcro	Health Care Related Organization	LOCATION_OF	diap	Diagnostic Procedure	23061
fndg	Finding	PROCESS_OF	humn	Human	20781
dsyn	Disease or Syndrome	PROCESS_OF	humn	Human	11115
bpoc	Body Part, Organ, or Organ Component	LOCATION_OF	fndg	Finding	9506
tisu	Tissue	LOCATION_OF	aapp	Amino Acid, Peptide, or Protein	9332
gngm	Gene or Genome	LOCATION_OF	genf	Genetic Function	9061
genf	Genetic Function	PROCESS_OF	gngm	Gene or Genome	8734
mobd	Mental or Behavioral Dysfunction	PROCESS_OF	humn	Human	7135
bpoc	Body Part, Organ, or Organ Component	LOCATION_OF	patf	Pathologic Function	6732
bpoc	Body Part, Organ, or Organ Component	LOCATION_OF	anab	Anatomical Abnormality	6377
lbpr	Laboratory Procedure	USES	lbpr	Laboratory Procedure	6187
aapp	Amino Acid, Peptide, or Protein	AUGMENTS	ortf	Organ or Tissue Function	6134
lbpr	Laboratory Procedure	USES	aapp	Amino Acid, Peptide, or Protein	5792
bpoc	Body Part, Organ, or Organ Component	LOCATION_OF	aapp	Amino Acid, Peptide, or Protein	5454
blor	Body Location or Region	LOCATION_OF	fndg	Finding	5253
topp	Therapeutic or Preventive Procedure	TREATS	dsyn	Disease or Syndrome	4388
dsyn	Disease or Syndrome	PROCESS_OF	mamm	Mammal	4315
bpoc	Body Part, Organ, or Organ Component	LOCATION_OF	dsyn	Disease or Syndrome	4149

[erikyao]

All hcro replacements (w/o semtype matching)

CUI1	concept_name1	semantic_type_abbreviation1	CUI2	concept_name2	semantic_type_abbreviation2
C1552516	Specialty Group	hcro	C0220961	UMLS Metathesaurus	inpr
C1516172	Cancer Center	hcro	C1513817	NCI-Designated Cancer Center	hcro
C0237680	Residential Care Institutions	hcro	C0035186	Residential Facilities	hcro
C0237680	Residential Care Institutions	hcro	C0035186	Residential Facilities	mnob
C1609437	Primary care clinic	hcro	C1552443	Clinic / Center - Primary Care	mnob
C1609437	Primary care clinic	hcro	C1552443	Clinic / Center - Primary Care	hcro
C0872261	repository	hcro	C3847505	Repository	mnob
C1306377	Postoperative anesthesia care unit	hcro	C0034871	Recovery Room	hcro
C1306377	Postoperative anesthesia care unit	hcro	C0034871	Recovery Room	mnob
C1552447	radiology facility	hcro	C1610162	Radiology Clinic/Center	mnob
C1552447	radiology facility	hcro	C1610162	Radiology Clinic/Center	hcro
C0013967	Emergency Service, Hospital	hcro	C0562508	Accident and Emergency department	hcro
C0013967	Emergency Service, Hospital	hcro	C0562508	Accident and Emergency department	mnob
C0338036	Doctor's office	hcro	C0031834	Physicians' Offices	hcro
C0338036	Doctor's office	hcro	C0031834	Physicians' Offices	mnob
C1546895	GlaxoSmithKline	hcro	C1552903	SmithKline Beecham	hcro
C1619637	Hospital Psychiatric Units	hcro	C0870667	Psychiatric hospital unit	hcro
C1619637	Hospital Psychiatric Units	hcro	C0870667	Psychiatric hospital unit	mnob
C1546882	NABI	hcro	C1552896	NABI	hcro
C1546858	Abbott Laboratories	hcro	C1552881	Abbott Laboratories	hcro
C1546873	Merieux	hcro	C1552891	Merieux	hcro
C1512798	Institute for Cancer Prevention	hcro	C1140168	NCI Thesaurus	inpr
C1546884	Novartis Pharmaceutical Corporation	hcro	C1552897	Novartis Pharmaceutical Corporation	hcro
C4699045	Stroke Center	hcro	C1136323	Logical Observation Identifiers Names and Codes	inpr
C3845566	Rehabilitation facility	hcro	C0034993	Rehabilitation Centers	hcro
C3845566	Rehabilitation facility	hcro	C0034993	Rehabilitation Centers	mnob

[andrewsu]

Great, this all looks fine to move forward. Please proceed with the update!

[erikyao]

Thank you for the confirmation!