How are the target - disease associations propagated in the ontology tree?

How are associations and evidence propagated in the ontology tree? There is a note in the documentation that RNA expression type evidence is not propagated, but how does it work for other types? Specifically, I was searching for the target VAC14 and looking at the associated diseases “striatonigral degeneration, childhood-onset”, and its’ parent term “genetic disorder”. It seems “genetic disorder” gets associations to VAC14 through several child terms, so it has more evidence than “striatonigral degeneration, childhood-onset”, however the overall score with VAC14 is lower. Is there a documentation page explaining the propagation?

Hi @Hilda! Thank you for your question, and welcome to the Open Targets Community :tada:

The propagation of evidence is covered in this section of the documentation: Target - disease associations - Open Targets Platform Documentation

Please let me know if you have further questions!

To answer this:

Evidence pages for a specific target-disease pair will display both direct and indirect evidence if it is available. (The VAC14 and Striatonigral Degeneration, Childhood-onset evidence page does not have indirect evidence because Striatonigral Degeneration, Childhood-onset does not have child terms in our ontology.)

However, when you look for diseases associated with a target — in your case, when you searched for VAC14 and viewed all the associated diseases in our associations page — the Platform will only use evidence from direct associations to calculate the association score. This is why “striatonigral degeneration, childhood-onset” has a higher score than “genetic disorder”.

Thanks a lot for the answer. Am I correct that when searching for a disease term (looking at the page for targets associated with a disease), for example “synucleinopathy”, it is only the indirect score that is presented? Does OpenTarget have a standardized way of combining the two types of evidence scores?

As an example, VAC14 is listed as top overall (indirect) score (0.73) for synucleinopathy. When looking into the details, this relation has only indirect- and no direct score. SNCA is listed as second overall score (0.72). However this relation also has a direct score of 0.11, so when ranking targets for this disease, one could argue that SNCA is actually more relevant than VAC14 due to it’s combination of direct and indirect score. Does OpenTargets have a view/summary/formula that shows relations based on a combination of the two score types?

I am also a little confused about the difference, there seem to be a lot of relations that have exactly the same direct and indirect score. For example “cystic fibrosis” and CFTR. When looking at the evidence page, I only see evidence for “cystic fibrosis” so I don’t understand where the indirect score comes from?

Hi @Hilda,

Good questions!

You are correct: when the disease term is fixed (you are viewing a list of targets associated with a disease), the targets are sorted based on the indirect score.

One thing that might be useful to you: using the API, you can modify this to see a list of targets associated with a disease, sorted by direct evidence only by using enableIndirect:false (see example query at the end).

Does OpenTarget have a standardized way of combining the two types of evidence scores?

The two types of scores are calculated using the same formula, which is a harmonic sum. The only difference is in which evidence is included: indirect associations scores reflect potential additional evidence from child terms of the disease.

In other words, if there is no indirect evidence for an association, then the direct and indirect association scores are the same. This is the case for CFTR and cystic fibrosis. Cystic fibrosis has no child terms in our ontology, so only direct evidence is available regardless of the view. You will see that the indirect and direct associations scores are the same (0.90).

Note, direct and indirect evidence are weighted the same in the overall calculation (in our scoring, direct evidence is ‘worth’ the same as indirect evidence).

Does OpenTargets have a view/summary/formula that shows relations based on a combination of the two score types?

We don’t, but I would be very curious to know what kind of view you have in mind!

Let me know if you have follow up questions :smiley:

Example Query for direct evidence

query targetsAssociatedWithPsoriasis {
  disease(efoId: "EFO_0000676") {
    associatedTargets(enableIndirect: false) 
      rows {
        target {