Run read-only SQL over the whole BRaVa gene-level results table.

61.8 million rows: every gene x trait x variant-mask x MAF-cutoff x ancestry cell, with the Burden, SKAT and SKAT-O p-values, the effect size and its standard error, and the cross-cohort heterogeneity test. Local, so a query costs no network.

Call schema() first. It returns the tables, the columns, worked query templates, and the semantic traps that make a syntactically valid query scientifically wrong here. Several of them invert the answer rather than degrade it.

Query the results view rather than the raw tables: it exposes p-values instead of -log10, and labels instead of integer codes, so the obvious query is also the correct one.

The tables, the query templates, and the traps. Read this before querying.

Returns the shipped tables with their columns and row counts, worked queries for the questions people actually ask, the analysis vocabulary (masks, MAF cutoffs, tests, significance thresholds), and a list of ways a correct-looking query gives a wrong answer on this data. That list is not boilerplate: it covers effect sizes that belong to a different test than the p-value beside them, a mask that is a calibration control rather than a biological category, ancestry strata that overlap, and p-values of exactly zero that mean the most significant result rather than a missing one.

Returns: tables, columns, recipes, vocabulary, thresholds and pitfalls.

Does a gene-trait association replicate across ancestries and biobanks?

BRaVa's distinctive view, and a tool rather than a documented query because the concordance count has to exclude the two pooled strata ('All' and 'non_EUR') that contain the same individuals as the ones being counted. The obvious SQL double-counts and looks entirely reasonable.

Pass a comma-separated list to screen a whole hit list at once: one gene at a time costs a call each, the list form returns a verdict per gene. Verdicts separate "underpowered" from "discordant", which is the distinction that matters when a stratum is fifteen times smaller than another.

Single-variant results for a trait, genome-wide or inside one gene.

Drops below the gene-level burden tests to the individual variants carrying a signal. Without gene this ranks the whole genome for the trait; with gene it restricts to that gene and adds the per-biobank effect-direction tally, the cross-biobank replication evidence.

Still fetched over HTTP rather than shipped in the database: the variant-level format is a separate, actively changing upstream release, an order of magnitude larger than the gene-level table, and rebuilt often enough that a local copy would be stale within the week. Each file is cached permanently once fetched.

Each row links to gnomAD, where population allele frequencies live.