How does the Platform display direct and indirect evidence?

Hello Open Targets Community !

Looking for overall association score between MKS1 (ENSG00000011143) and neoplasm (EFO_0000616), I found these results in the interface :

But when I click on it I can see some evidences, which were not mentionned in this first page :

Why are the evidences in Animal Models and Genetic associations not mentionned ?
ALso, I checked in the database, in table associationByDataTypeIndirect, and these evidences are also mentionned :

image

Can you please explain ?

Thanks in advance !

Hi @njeanray,

Thanks for your question!

A particularity of the Platform’s interface is that it will display direct or indirect evidence depending on which page you are looking at.

In this case, when you look at the associations page for diseases and phenotypes associated with MKS1, the Platform displays only direct evidence for the association between MKS1 and neoplasm, i.e. where each of these entities is specifically referred to.

However, when you click on the cell, the evidence page you land on displays both direct and indirect evidence for the association. The Platform uses the properties of our disease ontology (EFO) to gather evidence for MKS1 in neoplasm and any of its ontological descendants. In the evidence page, the evidence from GEL PanelApp is for renal fibrosis, and the evidence from PhenoDigm refers to CHILD Syndrome, rather than the more general term ‘neoplasm’.

Since these are indirect pieces of evidence, they will appear in the associationByDataTypeIndirect table.

You can find more information about this in our documentation.

Let me know if this has answered your question!

Best wishes,

Helena

Hello @hcornu ,

Thanks for your fast answer, I got it !
So if I undersand well, is the assumption telling that “all the direct evidences for a given disease-target association are included in indirect evidences for this disease-target association” correct ?

Does it mean that, somewhere, the score for the direct association is included in the indirect association score ?

Hi Nathalie!

Yes, that is correct. Each piece of evidence we have is given a single score, and the switch between ‘direct’ and ‘indirect’ only affects the evidence which is selected to be included in the aggregated scores (e.g. overall score or Data Type score).

When you view diseases associated with a target (for example, diseases associated with MKS1), the Platform will display a score which aggregates only direct pieces of evidence. In this example, the association between MKS1 and neoplasm has an overall score of 0.00 because the only direct evidence is from text mining.

However, when you search for targets associated with a disease (for example, targets associated with neoplasm), the Platform will display a score which aggregates both direct and indirect pieces of evidence. In this example, the association between neoplasm and MKS1 has an overall score of 0.53, because the score includes indirect evidence, e.g. from the Genomics England PanelApp.

Cheers,

Helena

Hello! Thanks for incorporating ontology into targets! As @njeanray mentioned, if I understood correctly, all targets that are targets of any of the descendants of an umbrella disease would be a regarded as a target of the umbrella disease itself, when quering the umbrella disease, e.g. Targets associated with hematopoietic and lymphoid system neoplasm the umbrella disease, and T-cell non-Hodgkin lymphoma profile page the descendant disease.

However, I downloaded the .tsv files of both diseases and found that there exists several targets of the T-cell non-Hodgkin lymphoma that are not the targets of the umbrella disease:
{‘APLP1’,
‘APOBEC2’,
‘ASPH’,
‘ECSIT’,
‘GAS1’,
‘GGTLC3’,
‘H4C13’,
‘IGHV4-4’,
‘MIR135B’,
‘PARD6A’,
‘RPL17-C18orf32’,
‘SAP18’,
‘SLC16A4’,
‘SNORD25’,
‘TRBV7-9’,
‘UPK1B’}.

Could you help me check if this is indeed the case, or is my understanding incorrect?

Thanks!

Hi @jasperhyp!

That is correct. When looking at targets associated with a disease, the Platform will also display indirect associations; that is, targets associated with diseases that are child terms of the disease in question.

Where did you download the .tsv files from? I tested several of the targets you mentioned by searching the targets associated with hematopoietic and lymphoid system neoplasm and the targets associated with T-cell non-Hodgkin lymphoma, and I was able to find them – see below an example for SAP18:


1 Like

Ah I think I got why. I downloaded the .tsv of hematopoietic and lymphoid system neoplasm (umbrella disease) from the exact same disease association page in March 2022 (two months ago), and the .tsv of T-cell non-Hodgkin lymphoma yesterday. It seems there has been a version update between this two time points, and the number of targets of the umbrella disease decreased from 10679 to 10350.

Sorry for bothering you… I just can’t download the large table of the umbrella disease now (because of an issue), but weirdly, I can download the T-cell non-Hodgkin lymphoma one (perhaps because it’s small), so I wanted to use these two datasets together. But it looks like there are indeed some inconsistencies (update) between the two versions.

No worries! We’re not entirely sure why the issue is happening for some diseases and not others, which is why I could only check by searching the table in the web app.

We did indeed have a new release recently, 22.04 – more info on the blog: Open Targets Platform 22.04 has been released! and Community: Open Targets Platform 22.04 is out now!

Let me know what inconsistencies you find! :grinning:

1 Like