AI-agent-guided pharmacogenomic analysis of the Swedish population
An open, governed access layer to the SweGen pharmacogenomic pilot dataset — 20,179 variants across 12 chromosomes with population allele frequencies in 1,000 Swedish individuals. Connect your favourite AI agent and run aggregate analyses; a Guardian agent reviews every query and every result against the data owner's sensitivity contract.
Read before you connect your agent
This is a public test of zero-trust data serving. Other approved users
share the same Jupyter kernel — namespace your variables. Do not
paste sensitive data into your agent prompts: anything you type may end
up in the audit log and may be visible to portal admins. Be aware that
prompt-injection content retrieved by your agent could cause it to
issue requests you did not intend. The Guardian is a defence-in-depth
layer, not a guarantee.
How it works
- Sign in with Hypha. Use your existing Hypha account, or register at hypha.aicell.io first.
- Request access. Tell the portal admins who you are and what you want to study. Adam Ameur (data owner) approves new users.
- Start a session. Once approved, click New session on your dashboard. The portal gives you a unique service URL.
- Paste the URL into your agent. Any agent that can fetch
SKILL.mdover HTTP and callrun_codecan use the portal — Claude, Cursor, your own agent, etc. The Guardian checks every code submission against the sensitivity contract; only reviewed outputs are returned.
The dataset
SweGen pharmacogenomic pilot. Pilot extract of Ameur et al. (2017) restricted to pharmacogenomically relevant loci with alternate-allele frequency ≥ 1%. Aggregated allele frequencies only — no individual-level genotypes, no per-sample data, no PII.
| Variants | 20,179 |
|---|---|
| Chromosomes | chr1, chr2, chr4, chr6, chr7, chr8, chr10, chr12, chr13, chr16, chr18, chr19 |
| Schema | chrom, pos, id (rsID), ref, alt, af |
| Reference build | GRCh38 |
| Data owner | Adam Ameur, Uppsala University / SciLifeLab |
| Sensitivity contract | aggregate AF analyses allowed; data exfiltration, host probing and secret extraction blocked |
Suggested analyses
The Safe Colab paper (Figure 3) demonstrates the kinds of analyses this portal supports:
- Allele-frequency spectrum stratified by variant type (SNV / insertion / deletion)
- Transition-to-transversion ratio per chromosome (expected ≈ 2.0)
- Six-class substitution spectrum (C>A, C>G, C>T, T>A, T>C, T>G — C>T dominates)
- Per-chromosome AF density
- Pharmacogene-level grouping (join against ClinVar / PharmGKB / CPIC star alleles you load yourself)
Get started
Loading…