Scientific validation of the release data

[addramir]

We need to figure out the plan how we scientifically validate the data we produce. @xyg123 could you please add the summary we discussed yesterday?

[xyg123]

Sure, here's the notes I took yesterday:

Gentropy Benchmarking

Additional QC + Benchmarking Steps

• Go through individual output columns and ensure constraints are met.

Study Index

• Duplicated studyIds.

Fine Mapping

• Sum of posteriors < 1.
• Direction of effect, formatting, etc.

Colocalisation

• Add validation for COLOC: h4/clpp between 0-1.

L2G

• Score between 0-1.

• Enrichment of drug targets.

• Density plot of score distribution.

• Interesting stories, gene targets, double check they are being picked up by L2G.

• Can we use the ready-curated gene lists in the AMP?

All Outputs

• Correlation with previous release?

• Consider a set of studies to refer back to (e.g. IBD) and cross-compare between releases.

• Can we identify shared top hits/credible sets between Gentropy and genetics portal?

  • Do we have resources to do this?

• Where are the discrepancies? (e.g. beta directions)

• Some of these issues already existed in the genetics portal.

  • Yakov and Prashant to go through community page and GitHub issues to see what can be addressed or closed.

Additional Considerations

• Consider a publication letter to address the issues with working with curated top-hits: “it’s ruining my pipelines 🙁”

• Ask Tobi to look at some studyLocusIds related to IBD.

[addramir]

As discussed with @xyg123 and @project-defiant: 1) Implement the logic of direct and indirect associations after evidence is calculated 2) Run enrichment on indirect associations for old OTG and new associations 3) Compare old and new assosiatnos