Hi K-dense team,
Love the work you guys are doing and would like to contribute!
I have an array of skills that relate to statistical genetics - think target discovery, validation, prioritization. I'd like to contribute the first one which is about having a pair of gene -> outcome link from coloclization which is the workhorse of genetic association that link target to disease.
The skill will generate the canonical locus compare sub plots, e.g. for sort1 liver eQTL and triglycerides/ldl:
However, I'm not sure what's the best way to contribute it to kdense.
the way i have it set up is that it has four primitives and the locus compare is orchasting them:
- get the outcome data (skill: gwas_catalog_region_fetch) - this is a fetch data skill from the gwas catalog
- get the exposure data (skill: eqtl_catalogue_region_fetch) - this is a fetch data from eQTL catalog
- compute LD (skill: ld-1000g-region-compute) - this is a fetch from 1000 genome (the right ethnicity for the data, default eur) and compute LD between sentinal coloc snp and region
- plot the locus compare (skill: locuscompare_region_render) - Harmonise + render + emit — joins exposure × outcome on variant_id, flips beta_outcome for swapped alleles, excludes palindromic A/T·G/C variants (count surfaced), then renders the 4 panels and emits PNG (like above).
I can ship it as the above but I noticed that k-dense has a database-lookup skill that consolidates lookups. I like the design. Should I start a database-fetch with 1 and 2 above to prime it (i can add more for other databases but the two above are the minimum). All of the skills I developed include unit tests as well as end to end examples that serve as golden parity to make sure code doesn't regress.
What do you think? How can I contribute it in the best manner?
Thanks,
Aviv
Hi K-dense team,
Love the work you guys are doing and would like to contribute!
I have an array of skills that relate to statistical genetics - think target discovery, validation, prioritization. I'd like to contribute the first one which is about having a pair of gene -> outcome link from coloclization which is the workhorse of genetic association that link target to disease.
The skill will generate the canonical locus compare sub plots, e.g. for sort1 liver eQTL and triglycerides/ldl:
However, I'm not sure what's the best way to contribute it to kdense.
the way i have it set up is that it has four primitives and the locus compare is orchasting them:
I can ship it as the above but I noticed that k-dense has a database-lookup skill that consolidates lookups. I like the design. Should I start a database-fetch with 1 and 2 above to prime it (i can add more for other databases but the two above are the minimum). All of the skills I developed include unit tests as well as end to end examples that serve as golden parity to make sure code doesn't regress.
What do you think? How can I contribute it in the best manner?
Thanks,
Aviv