Removing splicing QTL effect sizes to build "variantFunctionalConsequenceFromQtlId" evidence column.

[Juanmaria-rr]

Background

The current interpretation of splicing QTL (sQTL) beta coefficients for the direction of effect assessment is not informative, where negative are interpreted as loss of function and positive ones as gain of function.

For now we propose to remove them from the assessment for the direction of effect (DoE), so we will need to filter them out when building the evidences.

This issue is also addressing the point 1 of the issue raised by @Daniel-Considine here.

Tasks

• [ ] Filter out sQTL signals before getting the one with the biggest effect (get_biggest_effect function) in the generation of the variantFunctionalConsequenceFromQtlId evidence column: https://github.com/opentargets/evidence_datasource_parsers/blob/ea3008a6143fd1347a3629d5869fd50438004abc/modules/GeneticsPortal.py#L84
• [ ] Run the implementation of DoE in the backend with the new data to check that the direction of effect data is properly shown. *(Please note that I'm not familiar with the whole process of the evidence generation, so feel free to modify my suggested tasks)

Acceptance tests

When direction of effect data shows no directionality from sQTL evidences.

How do we know the task is complete?

1. Using an user case raised in genetics slack channel: On the associations page for Lymphocyte count associated with TYK2 (link), for the variant 19_10351837_C_T, the DoE is loss of function | Risk, where the loss of function comes from the Decreased gene product level derived from the highest effect size of sQTLs. After removing sQTL, this should appear as a gain of function | Risk.

[ireneisdoomed]

Looking at the ETL code, the value for variantEffect is based on variantFunctionalConsequenceFromQtlId, a value generated during the OTG evidence generation. @remo87 @Juanmaria-rr Can you confirm this is true? If so, it's up to us to filter out sQTLs in the logic and we'd have it fixed by next release.

[Juanmaria-rr]

Yes @ireneisdoomed, I think that would be the step to include to remove the sQTL effect sizes from the DoE assessment.

[DSuveges]

It's here So, the data team can update the parsers by filtering out sQTLs from the aggregation.

[d0choa]

In #3282, @addramir and @Daniel-Considine suggested changing the logic of the aggregation. Instead of using the largest effect size, which could be problematic for poorly powered studies, use the lowest p-value. I would try to introduce this change in the same script as well.

@Juanmaria-rr will have some estimates on the impact of this change.

@buniello This will imply changing a tooltip and documentation.

[DSuveges]

The parser and the data is updated. The data is here:

gs://otar000-evidence_input/Genetics_portal/json/genetics-portal-evidence-2024-04-16.json.gz

Benchmark against old data

Questions:

• Has the number of evidence changed?
• Has the number of evidence with directionality assignment changed?
• How many evidence have different QTL effect?
• How many times did the directionality changed?

Answers:

• There's no change in the number of evidence: 811,749
• Verified that the schema of the two datasets is the same.
• The number of evidence with directionality assessment dropped 12% to 210,050, which means 25.8% of evidence have directionality.
• Then number of evidence, where the directionality assignment changed is 67,170 representing ~31% of all evidence with directionality.