Harmonise pharmacogenetics associations that refer to star alleles

ireneisdoomed commented 7 months ago

As a developer I want to integrate and harmonise star allele data from PharmGKB into the Platform because these are clinically relevant associations that our users expect to see in our widgets.

Background

Star alleles are a widely used notation within the pharmacogenetics field. For 23.12 we won't show this information because of the complexities of standardising this notation to our variant IDs. @apriltuesday has explored its complexities in this notebook. For the next iteration, we want to integrate them as part of the pharmacogenetics dataset.

Tasks

[ ] Consolidate strategy to harmonise star allele nomenclature
[ ] Validate the presence of key drug associations with the corresponding star alleles.

Acceptance tests

Ellie has shared with us some key drug associations that we should use to validate the data. As it can be seen, most of the key associations using star alleles are lost (see column In OT).

Gene	Chromosome	Position	Reference	Alternative	RsID	Legacy Name	Alleles	Star allele	Key drug association	In OT
DPYD	1	97450058	C	T	rs3918290	c.1905+1G>A	*2A	*2A	capecitabine/fluorouracil	Yes
DPYD	1	97515787	A	C	rs55886062	c.1679T>G	*13	*13	capecitabine/fluorouracil	Yes
DPYD	1	97573863	C	T	rs56038477	c.1236G>A	HapB3	HapB3	capecitabine/fluorouracil	Yes
DPYD	1	97579893	G	C	rs75017182	c.1129-5923C>G	HapB3	HapB3	capecitabine/fluorouracil	Yes
UGT1A1	2	233760233	C	CAT	rs3064744	NC_000002.11:g.234668881TA[6]>TA[7]	*28	*28	Irinotecan	No
UGT1A1	2	233760498	G	A	rs4148323	211G>A	*6	*6	Irinotecan	No
TPMT	6	18130687	T	C	rs1142345	719A>G	*3C	3A/3C	thiopurine/mercaptopurine/azathioprine	No
TPMT	6	18130781	C	T	rs1800584	626-1G>A	*4	*4	thiopurine/mercaptopurine/azathioprine	No
TPMT	6	18138997	C	T	rs1800460	460G>A	*3B	3A/3B	thiopurine/mercaptopurine/azathioprine	No
CYP2D6					rs3892097		*4		codeine	No
CYP2C19					rs28399504		*4		clopidogrel	No
HLA					any variants?				abacavir/allopurinol	No
CYP2C9					rs1799853		*2		phenytoin/warfarin	No

Context

Metrics extracted by @apriltuesday and based on the 2023-10-05 data dump:

Total clinical annotations: 5073
        With RS: 4477 (88.25%)
                1. Exploded by allele: 13497 (3.0x)
                2. Exploded by drug: 18830 (1.4x)
                3. Exploded by phenotype: 23086 (1.2x)
Total evidence strings: 24641
        With CHEBI: 20451 (83.00%)
        With EFO phenotype: 8139 (33.03%)
        With functional consequence: 15633 (63.44%)
        With VEP gene: 15633 (63.44%)   
Gene comparisons per annotation
        With PGKB genes: 4220 (83.19%)  
        With VEP genes: 4099 (80.80%)   
        PGKB genes != VEP genes: 770 (15.18%)
Total RS: 2794
        With parsed alleles: 2771 (99.18%)
                With >2 alleles: 31 (1.12%)

ireneisdoomed commented 5 months ago

EVA will update the Pharmacogenetics dataset they provide to include star allele/gene associations so that we can display them in the widget. Our next meeting with them is on Jan 22nd.

For reference, the genes of the current pharmacogenetics dataset are the ones assigned by VEP. So the implicated variant requires to be normalised so that VEP can consume it. This is not possible for star alleles, however PharmGKB does have gene information. The intention is to show these associations as is, without the extra annotation from VEP.

ireneisdoomed commented 5 months ago

Useful issue tracking the investigations made by @apriltuesday: https://github.com/EBIvariation/opentargets-pharmgkb/issues/28

ireneisdoomed commented 5 months ago

After the discussion with EVA today, we have agreed that we want to maintain the star allele notation because it's well established in the field. This means that we need to introduce a new ID field in the schema to accommodate this. I propose haplotypeId. In the end, we'll have:

haplotypeId: for example CYP2C9*3, the name of the haplotype that comprises multiple variants.
variantRId: one of the variants in that haplotype, for example rs1057910. This information will come from PharmGKB, so that we can extract the most severe consequence.
genotypeId: each of the possible alleles in the variantId, for example 10_94981296_A_C

emcdonagh-OT commented 5 months ago

Could the haplotypeId hyperlink to the PharmGKB annotation for the * allele, therefore providing mapping to the contributing variants via this resource?
I don't know if we need variantRId or genotypeId here as I think most of the annotations are based on a single allele...is it possible to leave blank?
Would it be possible to also bring in the functional annotation e.g. no function, decreased function, normal function? This could be really useful for these annotations but also for example for applying to direction of effect annotations...maybe this is an additional feature down the road and could be a new widget under the variant consequence? Example: https://www.pharmgkb.org/clinicalAnnotation/982047500

apriltuesday commented 5 months ago

Hello! Some thoughts on these questions (and one question of my own)...

Linking to the haplotype page is surprisingly tricky as I think we'd need to find the PGKB ID for the haplotype (e.g. PA165816542, from which you can construct the URL), and this ID isn't in the data tables I'm currently using. I will look into it and keep you posted.
variantRsId and genotypeId can (actually must) be left blank here, we just need to remove the variant ID from the required fields in the schema...
We can get the allele function annotations (decreased function etc.), the annotation is technically available for all records but in my experience is only present for these named alleles. In theory the only possible values are documented in Table 2 here, but in a quick glance I see some variability in the PGKB data, so for now I would make this a generic string field - maybe alleleFunction?

On a separate note, for the genotype field - we currently fill this with the first column in PGKB's annotation table (rsid example, star example), which is labelled Allele but can be either allele or genotype. We can continue to do this for named alleles, as in both cases it would document the raw text that we use to create the genotype OR haplotype ID, or we could do something else. Do you guys use this field at all? Do you have a preference?

apriltuesday commented 5 months ago

Hi @emcdonagh-OT @ireneisdoomed @DSuveges - Getting the information to link to a haplotype page from the haplotypeId will not be possible in general, as sometimes the annotation is actually for a genotype and not a haplotype at all... (example here).

However it should be achievable for the majority of cases, so perhaps can be added as a non-required field. This would be PGKB's internal identifier for the haplotype - for example (field name pending):

{
    ...
    "haplotypeId": "CYP2C9*1",
    "internalHaplotypeId": "PA165816542"
    ...
}

from which you can get the URL as https://www.pharmgkb.org/haplotype/PA165816542.

Does this seem reasonable?

ireneisdoomed commented 5 months ago

@apriltuesday Apologies for the late reply, I'd missed this. That is absolutely reasonable, we expect that haplotypes won't be available for many evidence. Using this ID to link back to PharmGKB is a nice to have, but not essential since we already link the whole evidence. Is it complex to extract? As for the field name, I'd suggest haplotypeFromSourceId. Let me know if this is OK and we'll update our schema.

apriltuesday commented 5 months ago

@ireneisdoomed thanks, haplotypeFromSourceId sounds good, it's not complicated to extract for the majority of cases so should be fine.

The other schema changes I think are needed are:

variantRsId needs to be not required - you could make it so that either variantRsId or haplotypeId are required though
alleleFunction (or another name) to be added - if you want the "decreased function" etc. annotation to be included in this submission

ireneisdoomed commented 5 months ago

Sorry for the delay @apriltuesday We have a new schema version under the tag 2.6.1 with the changes you have requested:

variantRsId is now optional (all IDs are now optional)
haplotypeFromSourceId will contain PGKB's haplotype ID to link to
directionality contains the allele function annotation

And one more:

drugFromSourceId will contain the CHEBI identifiers. This used to be stored in drugId, but we want to leave this field blank. Is this possible?

Let me know if you have any issues, thank you so much!

apriltuesday commented 5 months ago

Great, thank you! I'll make the necessary changes and let you know any issues.

ireneisdoomed commented 5 months ago

The latest EVA submission (cttv012-2024-02-02_pgkb.json.gz) contain the evidence from the haplotypes in the table above.

Gene	Chromosome	Position	Reference	Alternative	RsID	Legacy Name	Alleles	Star allele	Key drug association	In OT 23.12	In OT 24.02
DPYD	1	97450058	C	T	rs3918290	c.1905+1G>A	*2A	*2A	capecitabine/fluorouracil	Yes	Yes
DPYD	1	97515787	A	C	rs55886062	c.1679T>G	*13	*13	capecitabine/fluorouracil	Yes	Yes
DPYD	1	97573863	C	T	rs56038477	c.1236G>A	HapB3	HapB3	capecitabine/fluorouracil	Yes	Yes
DPYD	1	97579893	G	C	rs75017182	c.1129-5923C>G	HapB3	HapB3	capecitabine/fluorouracil	Yes	Yes
UGT1A1	2	233760233	C	CAT	rs3064744	NC_000002.11:g.234668881TA[6]>TA[7]	*28	*28	Irinotecan	No	Yes (1451206982)
UGT1A1	2	233760498	G	A	rs4148323	211G>A	*6	*6	Irinotecan	No	Yes (1451329460)
TPMT	6	18130687	T	C	rs1142345	719A>G	*3C	3A/3C	thiopurine/mercaptopurine/azathioprine	No	Yes (1451237326)
TPMT	6	18130781	C	T	rs1800584	626-1G>A	*4	*4	thiopurine/mercaptopurine/azathioprine	No	Yes (1184648909)
TPMT	6	18138997	C	T	rs1800460	460G>A	*3B	3A/3B	thiopurine/mercaptopurine/azathioprine	No	Yes (1451237326)
CYP2D6					rs3892097		*4		codeine	No	Yes (1183616718)
CYP2C19					rs28399504		*4		clopidogrel	No	Yes (1043858794)
HLA					any variants?				abacavir/allopurinol	No	Yes (981419257/981419260)
CYP2C9					rs1799853		*2		phenytoin/warfarin	No	Yes (1447672988/982047500)

This is an example of how CYP2C9*2 and warfarin is represented:

-RECORD 0----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 datasourceId                   | pharmgkb                                                                                                                                                                                                                                                                                                                                                                                                                                                                               
 datasourceVersion              | 2024-01-05                                                                                                                                                                                                                                                                                                                                                                                                                                                                             
 datatypeId                     | clinical_annotation                                                                                                                                                                                                                                                                                                                                                                                                                                                                    
 directionality                 | Decreased function                                                                                                                                                                                                                                                                                                                                                                                                                                                                     
 drugFromSource                 | warfarin                                                                                                                                                                                                                                                                                                                                                                                                                                                                               
 drugFromSourceId               | CHEBI_10033                                                                                                                                                                                                                                                                                                                                                                                                                                                                            
 evidenceLevel                  | 1A                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     
 genotype                       | *2                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     
 genotypeAnnotationText         | The CYP2C9*2 allele is assigned as a decreased function allele by CPIC. Patients with CYP2C9*2 in combination with another normal function allele, a decreased function allele, or a no function allele may have increased risk of bleeding when treated with warfarin as compared to patients with two normal function alleles. However, conflicting evidence has been reported. Other genetic and clinical factors may also influence the risk of warfarin-induced bleeding.         
 genotypeId                     | NULL                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   
 haplotypeFromSourceId          | PA165816543                                                                                                                                                                                                                                                                                                                                                                                                                                                                            
 haplotypeId                    | CYP2C9*2                                                                                                                                                                                                                                                                                                                                                                                                                                                                               
 literature                     | [29432897, 10509530, 31186542, 29432897, 17653141, 17653141, 17653141, 17653141, 17653141, 18756910, 24503627, 24503627, 11926893, 10509530, 18574025, 24503627, 24503627, 24474498, 25001883, 25001883, 23932037, 23932037, 23774101, 23104259, 23104259, 25521356, 25521356, 24602049, 25769357, 25858232, 14676821, 17955230, 18690342, 22571356, 19297219, 23602689, 21148049, 25244877, 15714076, 23348161, 18570163, 26777610, 27488176, 27581200, 28033245, 28033245, 28689179] 
 pgxCategory                    | toxicity                                                                                                                                                                                                                                                                                                                                                                                                                                                                               
 phenotypeFromSourceId          | MP_0001914                                                                                                                                                                                                                                                                                                                                                                                                                                                                             
 phenotypeText                  | Hemorrhage                                                                                                                                                                                                                                                                                                                                                                                                                                                                             
 studyId                        | 1447672988                                                                                                                                                                                                                                                                                                                                                                                                                                                                             
 targetFromSourceId             | ENSG00000138109                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
 variantFunctionalConsequenceId | NULL                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   
 variantRsId                    | NULL

Unless there are any comments, I consider this work done. Thanks again @apriltuesday for your effort!

opentargets / issues