Questions about the Open Targets Platform (posterior probability, odds ratio for drug associations, preference for hypothesis-independent genetic data)

  • For some genetic markers the posterior probability is low but LG2
    score is high, they go opposite direction, how to interpret result?
  • I notice that sometimes when I look one variant and then
    other the part of molQTL remains unchanged so sometimes I do the mistake
    thinking that they have the same molQTLs. It is for both Safari and
    Chrome browsers. Then for some markers the molQTLs referring to the
    eQTLs do not appears even if I know that they are but for other
    variants the molQTL deriving from eQTLs appears but pQTLs not. It
    has
    happened that for one of my markers I saw one time tuQTLs but when I
    returned to the page they disappeared and now I see just one pQTL.
  • I have one question referring to the Open Targets old
    version, if
    for the drug association I see the Odds Ratio 2.0 it means that the
    drug work more efficiently?
  • Of course I see still benefits coming from the old version
    of Open
    Targets, and I really don’t know what I will do without it my
    work.
    I like very much the new Open Targes but the hypothesis independent genetic data are more discovery for me.

Thank you for all your work with the platform, I started many years ago with Immunobase and saw all the transformation in these years.

1 Like

Hi Magdalena,

Thank you for showing an interest!

I’d be happy to look into this - could you provide some examples to show what you mean?

Hi @Magdalena,

Thank you for reporting the second point — I have just checked with the team and this will be fixed in the next release (25.06).

I will find out about the rest of your questions :slight_smile:

Thanks,

Helena

  • For some genetic markers the posterior probability is low but LG2
    score is high, they go opposite direction, how to interpret result?

Regarding this point - the posterior probability for a genetic variant is the probability it is the causal variant for a given credible set.
The L2G score is separate - this is the prioritisation score for a gene with respect to the credible set.
Low posterior probabilities for variants can be caused by uncertainty in the fine-mapping, but this doesn’t prevent the ML model from assigning a high L2G score to that credible set if there are informative features.