DoseDNA — Ask your genome a question

Plain language on the surface. Hard clinical data beneath.

Behind the conversation sits a deterministic engine: your file → diplotype → phenotype → a CPIC gene-drug recommendation. Each answer is the output of that pipeline, phrased for you — not a model's opinion about your health.

get_gene_status

Reads your status

Returns your metabolizer phenotype for a specific gene from the variants the engine called — or “not determined” when the file lacks the deciding read.

lookup_cpic_recommendation

Fetches CPIC guidance

Queries CPIC’s live API for the verbatim recommendation, evidence classification, and implications text for your (gene, phenotype, drug) — no paraphrasing the underlying source.

check_drug_interactions

Catches interactions

Runs your medication list against bundled FDA/CPIC tables for drug–drug and phenoconversion flags — inhibitors and inducers that can shift your effective phenotype.

suggest_clinician_questions

Builds your question list

Generates 4–6 concrete questions to bring to your clinician or pharmacist, grounded in your phenotypes and the meds you mentioned this session.

One status. Two drugs. Opposite advice.

Your CYP450 enzymes process almost every prescription drug. The same slow enzyme helps one medication and harms another — so a single "normal/abnormal" label is dangerously incomplete.

ClopidogrelProdrug · must be switched on

↓ under-activated

A poor metabolizer activates too little of the drug. It may be less effective at preventing clots.

Effectiveness flag

EscitalopramActive drug · must be cleared

↑ under-cleared

The same slow enzyme clears the drug too slowly, so it can build up toward toxicity.

Accumulation flag

Same gene CYP2C19, same poor-metabolizer status — opposite actions. DoseDNA encodes the direction of risk, and never quietly rounds a missing result up to "normal."

An agent that knows what it isn't.

The most important thing a health agent does is refuse to overstep. These boundaries are wired in, not improvised.

✓

Explains the mechanismWhy a gene changes what a drug does, in language you can repeat to a clinician.

Never gives a doseNo numbers, no “take this much.” That decision belongs to your prescriber.

✓

Says “not determined” out loudIf your file lacks the deciding variant, it tells you — it never assumes normal.

Never tells you to stop a medIt hands you questions for a professional, not instructions to act on alone.

✓

Shows its workEvery answer is tied to a tool call you can see — no hidden reasoning about your health.

Never uploads your DNAOnly de-identified {gene, status, drug} leave this tab through any channel we observe — and each real query is mixed with 5 decoy queries to make casual log inspection harder. Browser-extension and OS-level data flows are out of scope.

Your raw DNA file stays in this tab.

Selecting a file is not uploading it. DoseDNA parses your data in a Web Worker on your device. The only thing that ever leaves is a minimal de-identified question — gene name, metabolizer status, drug name — and even that gets mixed with 5 decoy queries so the provider can't tell which one is yours.

What happens to…Where it goes

Your raw DNA filenever sent

Your rsIDs & genotypesnever sent

Your diplotypenever sent

{ gene, phenotype, drug } — 3 labelsmixed with 5 decoys

DNA bytes through observed channels0

Guidance is sourced live from CPIC — the same peer-reviewed gene-drug guidelines clinical pharmacists apply by hand.

Talk to your DNA — your raw file stays in this tab.

Load your 23andMe or AncestryDNA file and start a conversation. The file is read in this browser tab. Only minimal de-identified summaries (gene name, metabolizer status, drug name) are sent to the AI when you ask a question — never the raw genome.

Start chatting with DoseDNA

Informational only — not medical advice. Confirm everything with a qualified clinician or pharmacist.