Adding MR/ratio beta to colocalisation

[addramir]

Old OTG platform in colocalisation page there is a study beta: https://genetics-docs.opentargets.org/faqs#why-are-betas-and-odds-ratios-displayed-inconsistently-in-the-portal

On the other hand, we display the study beta coefficient in the colocalisation table of the study locus page e.g. [LDL cholesterol (GCST002222) with locus around 19_44886339_G_A (rs7254892)](https://genetics.opentargets.org/study-locus/GCST002222/19_44886339_G_A). This beta is with respect to the alternative allele of a single variant, the lead variant at the top of the study locus page (i.e. rs7254892 for the LDL cholesterol study).

The reason we have decided to display the study beta it to facilitate the comparison of the direction of effect across different colocalising tissues.

We need something similar for our colocalisation widget.

[Tobi1kenobi]

After discussing with @addramir, the issue seems to lie with the gentropy implementation of coloc not returning a way of interpreting direction of effect. This can be resolved by generating a ratio of the SuSiE posterior means (mu) between either the GWAS and GWAS or the GWAS and molQTL.

In the instance where the two lead variants are the same, this is trivial. When these do not align a decision must be made about what the most appropriate comparison is.

Actions:

• [x] Modify colocalisation.py to return the ratio of mus
• [x] Update the schema to include ratio of mus with appropriate name
• [x] Determine how to handle edge cases
• [x] Create a test dataset for edge cases to ensure nothing breaks downstream

[addramir]

Just as an idea: calculate the signed CLPP weighed on CLPP, similar to eCAVIAR but taking into account the sign of the betas. E.g.: -----|CS1 pip,beta|CS2 pip,beta snp1 |. 0.3,+1. | 0.6, +1 snp2 | 0.5, -1. | 0.3, -1 snp3 | 0.2, +1. | 0.1, -1

signed_CLPP=sum(sign_1 sign2 pip1 pip2)/(sum(pip1 pip2))=(0.3 1 0.6 1+0.5 (-1) 0.3 (-1)+0.2 1 0.1 (-1)) / (0.3 0.6+0.5 0.3+0.2 0.1)=0.8857

OR

-----|CS1 pip,beta|CS2 pip,beta snp1 |. 0.3,-1. | 0.6, +1 snp2 | 0.5, -1. | 0.3, +1 snp3 | 0.2, -1. | 0.1, +1

signed_CLPP=sum(sign_1 sign2 pip1 pip2)/(sum(pip1 pip2))=(0.3 (-1) 0.6 1+0.5 (-1) 0.3 1)+0.2 (-1) 0.1 (1)) / (0.3 0.6+0.5 0.3+0.2 0.1)=-1

[Tobi1kenobi]

Currently I've been working on a subset of the data and trying to determine how many credible sets contain variants which all have agreeing sign(beta1/beta2). Assuming most (90%+) sets have all variants in agreement, I was considering reporting, for the lead variant, the beta ratio or potentially more simply whether the eQTL effect is positive or negative relative to a positive GWAS effect (as this circumvents the choosing a variant issue and is all we really care about, right?).

I'm a little cautious about the above approach... I don't like the idea of a signed probability, personally, and whilst I can see the logic behind it, I'm afraid it may be a little confusing to explain/justify to a casual user of the portal to immediately interpret with a tooltip/FAQ.

[Tobi1kenobi]

Resolved with https://github.com/opentargets/gentropy/pull/854