Differences in overall association scoring between target-disease and disease-target

General
1

Hi all,

I am a little confused about why the overall association scoring is different between a case where I first search for a disease and then look for a target, and another case where I first search that same target and then look for that same disease as well. Perhaps an example would be a good to illustrate what I mean:

Target: PSEN1
Disease: Alzheimer’s Disease

a) If I search for PSEN1, and then look at its association with Alzheimer’s Disease, the overall association score is 0.44.

Screen Shot 2022-09-27 at 10.47.10 AM
Screen Shot 2022-09-27 at 10.47.10 AM941×740 70.7 KB

b) If I search for Alzheimer’s Disease, and then look at its association with PSEN1, the overall association score is 0.87.
(I can’t put another screenshot but here’s a link instead: Open Targets Platform)

May I know why there is a difference?

Thanks a bunch! :smiley:

2

Hi @keithzikai.chew!

The difference has to do with how we display evidence, as you can read in this other Community post: Score values from Disease->Target vs. Target->Disease

Essentially, if you search for diseases associated with a target (your first use case), the association score displayed is calculated using only direct evidence, i.e. in this case, evidence that links PSEN1 to Alzheimer’s disease specifically.

If you search for targets associated with a disease (your second use case), the association score calculation includes both direct and indirect evidence: evidence that links PSEN1 to Alzheimer’s disease and any of its child terms in our ontology such as Early-onset autosomal dominant Alzheimer disease (screenshot below from the Alzheimer’s disease profile page).

Ontology of Alzheimer's disease in the Open Targets Platform
Ontology of Alzheimer's disease in the Open Targets Platform2504×510 83.1 KB

This indirect evidence is also displayed in the evidence page for Alzheimer’s disease and PSEN1, see example below, where the ClinVar widget displays evidence for early-onset autosomal dominant Alzheimer disease.

ClinVar widget in the evidence page for Alzheimer's disease and PSEN1
ClinVar widget in the evidence page for Alzheimer's disease and PSEN12504×1130 472 KB

You can also read more about this in our documentation: Target - disease associations - Open Targets Platform Documentation

3

Hi @hcornu ,

This is a great explanation! Thank you so much :smiley:

Cheers,
Keith

1 Like
4

Hi all,

I do have a follow up question to the answer I got above from @hcornu . I am using GraphQL API to query the association score between PSEN1 and Alzheimer’s Disease. I notice that if my query is:

query target_disease($ensemblId: String!) {
  target(ensemblId: $ensemblId) {
    id
    associatedDiseases (
      enableIndirect: true
    ) {
      count
      rows {
        score
        disease {
          name
          id
        }
      }
    }
  }
}

then the score when I locate “Alzheimer’s Disease” is 0.9048851913021447. However, when my code is:

query disease_target($efoId: String!) {
  disease(efoId: $efoId) {
    name
    id
    associatedTargets (
      page: {index: 0, size: 50}
      enableIndirect: true
    ) {
      count
      rows {
      	score
        target {
          approvedSymbol
          id
        } 
      }
    }
  }
}

the score when I locate “PSEN1” is 0.8718348048253193 which is slightly different.

In other words, when asking for indirect evidence score between target-disease and disease-target, they give slightly different scores. May I ask why is this so?

(Unsurprisingly, when looking at direct evidence between target-disease and disease-target, I get the same score of 0.436218251231685. I just can’t figure out why indirect evidence would be scored differently between a target-disease query and a disease-target query.)

An explanation would be greatly appreciated :slight_smile:

In addition, may I ask if there is a way to quickly retrieve the overall association score between a target and disease from the GraphQL API? What I am doing now is searching for associatedTargets to a disease, and then iterating through the associatedTargets till I get the target I was searching for. Takes a long time for the API to fetch the score in this manner. A shorter method would be nice to know:) Sorry for the long questions :frowning:

Thanks!

Cheers,
Keith

1 Like