You upload a variant file. We hand back an annotated variant list you can query, analyse, and hand to the AI agent. This page is a short description of how that works and what you can rely on.
Engineering notes, not a contract
Running this offline
The pipeline takes a file with variant identifiers in it and returns the same rows, in the same order, with a large set of functional annotations attached to each one. Your original columns are preserved, the annotation columns come from the annotation database, and the result is a variant list you can query without a separate load step.
You do not need to tell it what format your file is in. It detects the format, the delimiter, the header row, and the identifier style by looking at the file.
Any tabular file with variant identifiers, plus a few scientific formats we recognise and parse specially.
| Format | What the pipeline does with it |
|---|---|
| GWAS summary stats | Recognises p values, effect sizes, and variant identifiers regardless of column naming. Your statistical columns are kept alongside the annotations. |
| Credible sets | Fine mapping output from tools like SuSiE, FINEMAP, or PAINTOR. Posterior probabilities and credible set assignments are preserved. |
| VCF files | Single or multi sample. Chromosome, position, reference, and alternate alleles are extracted. Relevant INFO fields are parsed. |
| Plain variant lists | Any TSV or CSV with a variant identifier column. rsID, VCF style, and internal identifiers can be mixed in the same file. |
A run goes through five stages. Each one has a single job, so a failure in one stage can be retried without redoing the others.
One of the ugliest parts of variant analysis is that the same variant has several names. The pipeline handles this on your behalf.
A variant list with your original columns on the left and the full annotation payload on the right. The annotation side covers the same categories documented in the Data and Annotations guide, which is the canonical reference for which columns ship in the current release.
Try it