01 · SOURCES

We don't make up findings.

Every line on the Medication Safety Card and every plain-English summary in the Twenty Rooms traces back to a curated, peer-reviewed pharmacogenomic or clinical-genomic database. We don't run novel research. We translate established evidence into something you and your doctor can act on in two minutes.

CPIC — Clinical Pharmacogenetics Implementation Consortium

Practice guidelines for drug–gene interactions, written by a global expert panel and updated continuously. We use CPIC Level A and Level B recommendations to drive every wallet-card line. If CPIC says a drug needs an alternative for your metabolizer phenotype, we surface it.

PRIMARY

PharmGKB

The broader pharmacogenomic knowledgebase that CPIC sits on top of. We use PharmGKB to populate the "why" copy on findings — variant annotations, allele frequencies, and supporting study citations — so a curious reader can verify a line all the way to a paper.

PRIMARY

ClinVar

NCBI's catalogue of variant–disease associations. We use ClinVar's "Pathogenic" and "Likely Pathogenic" classifications for hereditary-condition findings (BRCA1/2 founder variants, CFTR carrier status, etc.), and we treat anything below "Likely Pathogenic" as informational rather than clinical.

PRIMARY

SNPedia & published GWAS catalogs

For non-clinical traits and lower-evidence findings (caffeine metabolism, lactose tolerance, taste perception), we draw from SNPedia and the GWAS Catalog. These are shown in the body-system reports but never on the Safety Card.

SECONDARY

Evidence tiering is not cosmetic. A finding only reaches the Medication Safety Card if it has CPIC Level A or equivalent support. If we have a hunch that turns out to be a single-paper observation, you'll see it in the report — never on the card.

02 · THE PIPELINE

How a variant becomes a card line.

A consumer-chip raw file is roughly 650,000 single-nucleotide positions. Most of them aren't relevant to medication safety. The interesting work happens in five steps:

  1. Variant detection Your raw file is parsed against a reference panel. The engine identifies which positions match the reference, which are heterozygous, and which are homozygous variant. All on your machine — the file never leaves it.
  2. Star-allele calling For pharmacogenes (CYP2C19, CYP2D6, DPYD, TPMT, SLCO1B1, VKORC1, HLA-B, and others), we resolve the underlying variants into functional alleles — the named haplotypes that determine whether you metabolize a drug poorly, normally, or unusually fast. This is where "you carry CYP2C19 *2/*2" becomes "poor metabolizer."
  3. CPIC recommendation lookup For each phenotype, CPIC publishes drug-specific recommendations: avoid, alternative agent, alternative dose, monitor closely. We translate the matched recommendation into the card's "why" line ("won't work as a painkiller", "start at a lower dose, check INR on day 3", etc.).
  4. Severity-tier assignment Each finding is assigned three, two, or one pip based on the clinical consequence of getting it wrong. Anaphylaxis or treatment failure gets three; meaningfully altered dose or response gets two; useful-to-know-but-not-urgent gets one. The tiers are deterministic — same finding, same pip count, always.
  5. Plain-English composition Dr. Prime composes the readable copy from the structured finding, the recommendation, and the severity. No allele letters, no multipliers, no abbreviations make it to the user-facing text — that translation is the product.

03 · SEVERITY TIERS

Three dots, two dots, one dot.

The wallet card encodes severity by number of pips, never by colour. The brand is monochrome on purpose — a "red dot" reads as alarm regardless of context, and we don't want a routine "lower-dose starting point" to look like a code blue.

●●● Critical. Risk of a serious adverse event (anaphylaxis, severe cutaneous reaction, treatment-failure for a life-threatening indication) if the standard regimen is used without modification. Always discuss before any new prescription. ●●○ Caution. Standard regimen will likely under- or over-perform. Alternative dose or alternative agent is recommended; outcome is meaningfully altered, but not immediately dangerous. ●○○ Note. Useful context for a prescriber. May influence monitoring frequency or choice between equivalent agents. Not urgent, but worth mentioning.

Pips correspond to clinical consequence, not to how rare the variant is. A common variant with serious consequences gets three pips; a rare variant with mild consequences gets one.

04 · WHAT'S NOT IN THE REPORT

The deliberate gaps.

A lot of consumer-genomics products surface every variant they can recognize, in the name of "completeness." We don't, and the reasons are intentional:

Rare variants below population-frequency thresholds

If a variant appears in fewer than 1 in 50,000 people and the supporting evidence is a single case report, we don't surface it. The signal-to-noise ratio is too poor; the false-positive worry it would cause outweighs the rare true positive.

Polygenic risk scores without ancestry calibration

PRS distributions are population-specific. A score that's been calibrated only on European-ancestry reference panels can mis-rank everyone else by orders of magnitude. Where validated ancestry-calibrated PRS exists, we use it; where it doesn't, we mark the finding NOT_TESTED rather than misreport. We will not invent precision we don't have.

Findings without an actionable recommendation

If we can detect a variant but there's no CPIC-level recommendation that follows from it — and no clear clinical guidance from any equivalent body — it stays in the underlying data but doesn't generate a card line. We won't manufacture a recommendation to justify a finding.

Panels we haven't yet validated

The engine grows over time. New panels (new pharmacogenes, new carrier-screen conditions, new risk-stratification scores) are added as the evidence matures, not as the marketing demands. If a panel isn't validated against our internal benchmark, it's not in the report — even if competitors include it.

05 · BUILD & UPDATES

Reference build and update cadence.

Internally, the engine standardizes on GRCh38 as the reference build. If your raw file uses GRCh37, we lift over to GRCh38 automatically; the report tells you which build was used so a clinician can verify positions if needed.

Updates ship twice yearly with new CPIC guidelines, new validated panels, and corrections to prior interpretations. When an interpretation changes — and they sometimes do, as evidence accumulates — your previous reports remain readable and the engine flags exactly which finding changed and why. We will never silently downgrade or rewrite history.

Update mechanics are local: an updated engine ships as a new build of the desktop app, signed with our minisign release key. The update path verifies the signature before installing, and you can opt out and stay on a frozen version forever if you prefer.

06 · WHAT THIS IS NOT

The honest boundary.

  • Not a medical device. SynaptiX is informational software for use in conjunction with a licensed clinician. It does not diagnose, treat, or prescribe.
  • Not a diagnostic test. A negative finding is not a clinical clearance. A positive finding is not a diagnosis. Both are evidence to be weighed alongside everything else your doctor knows about you.
  • Not a substitute for clinical judgment. The card is a conversation starter, not an order. Your prescriber's experience, your full history, and current labs always take precedence.
  • Not an ancestry or "fun traits" report. If you want to know where your great-grandparents are from, or how your earwax type compares to a global average, this isn't the right product. SynaptiX is tightly focused on actionable medication safety and body-system summaries.
  • Not perfect. Pharmacogenomic evidence is a moving target. We update the engine on a cadence that lets us be confident, not loud. If an interpretation here contradicts a more recent guideline, that guideline wins — and we'll catch up at the next release.