Skip to content

just_dna_format.locus_genes

just_dna_format.locus_genes

The two study-locus tables: which gene a GWAS locus points at, and which molecular trait shares it (0.8, RM333).

locus_to_gene.csv and colocalisations.csv are derived-fact sidecars filled by just-dna-enricher's opentargets enrich pass from the Open Targets Platform snapshot lane. They are the inputs to an effector-gene judgement and never the judgement itself: the format does not author or compute a causal or effector confidence label (RM323's cut). A module that wants one states its rubric over these columns, and a corrected input goes in overrides.csv.

Both are keyed on the study locus, Open Targets' own grain, and on the module variant. A study locus is one credible set from one study (studyLocusId, a hash the source mints). One variant sits in several of them, one per study that fine-mapped the region, and each names its own candidate genes with its own score, so a row per variant would have to pick a study on the module's behalf. Several study loci per variant are kept and counted, never collapsed to a winner (@multiplicity-is-a-finding).

Members, not lead-only (decided 2026-11-04). A row exists for every credible set a module variant is a member of, whether or not it leads it: HFE C282Y (rs1800562) sits in 82 GWAS sets of the 26.09 release and leads none, so a lead-only table would be empty for the module the lane exists for. Two module variants in one set each get the set's rows, so variant_key is in both keys. The member's own weight in the set is its posterior on gwas_effects.csv (RM239), not a column here.

Why coloc is a second table and not columns on the first. A colocalisation pairs one GWAS study locus with one molecular-QTL study locus, and the gene and tissue sit on the right-hand study. The release measured 2026-11-04 (probes/LANES_0_8.md § built) has many right-hand loci per (study locus, gene) pair (one per tissue and per QTL kind), so h4/clpp on the L2G row would have to pick a tissue. The pair is kept at the source's own grain instead, and the gene it names is a column a consumer joins on.

variant_key is the module's own key for the member variant, the subject the pass asked about, so these rows join the module's weights directly. The source's own spelling of the set's lead (19_44908684_T_C, or an OTVAR_… hash for a long allele) is kept beside it as lead_variant_id for tracing back to the release, and is never parsed; it names another variant whenever the module's variant is a member and not the lead.

LocusToGeneRow

Bases: BaseModel

One (study locus, gene) pair and the locus-to-gene score the source gave it.

Standalone (not an AuthoredModel) for the reason every fact row is: machine-produced reference fact, not an authored annotation, with extra="forbid" so a typo'd column is caught.

ColocalisationRow

Bases: PairedFields

One colocalisation between a GWAS study locus the module's variant is a member of and a molecular-QTL study locus.