Direct-to-consumer genetic testing

Mail-back genetic testing compresses an entire clinical discipline into a plastic funnel, which makes understanding its physics less optional than for tests ordered through physicians. Two quite different reading technologies hide behind identical-looking reports, and the gap between them decides every downstream claim.

What each technology physically reads

Genotyping arrays pose a multiple-choice exam. Fixed probes across the surface interrogate several hundred thousand predefined positions — sites chosen years ago for their commonness or historical interest — reporting which letters sit there. Per-position accuracy is superb and cost minimal, but anything absent from the questionnaire remains permanently invisible: rare disease-causing variants, structural rearrangements and most of the genome’s true novelty simply cannot appear in results. Whole-genome sequencing instead re-reads all three billion positions afresh, typically thirty times over so that random errors wash out in consensus and genuine deviations survive. It costs more, misses its own repetitive corners, and sees what no array anticipated. One further shared complication arrives before either instrument runs: saliva contains substantial bacterial DNA alongside the human fraction, so extraction quality silently bounds how much human material ever reaches measurement.

The genetics these devices report on

Consumer findings sort into three species. Carrier results concern recessive conditions: possessing one broken copy causes no illness in the holder while informing reproduction — clean Mendelian facts, well suited to mail delivery. Actionable pharmacogenomic variants adjust doses of specific drugs. Everything else — heart attack, diabetes, depression, hypertension — belongs to the third and strangest category: polygenic risk, where hundreds to thousands of common spelling differences each nudge probability by a few percent. Polygenic scores sum those nudges using weights learned from enormous association studies, producing a lifetime-risk ranking among peers. The catch is geometrical rather than statistical: case and score distributions overlap so heavily that a person at the ninetieth percentile often has less absolute risk than someone at the median who smokes. Percentiles describe crowds; nobody’s body reads like a crowd.

Association-derived weights also carry an ancestry passport problem. Because discovery cohorts skew toward European-descent participants, scores transfer imperfectly to other populations whose linkage patterns and variant frequencies differ — an active research embarrassment that responsible providers disclose and careless ones omit.

Reading a report without being read by it

Three failure modes deserve deliberate inoculation. A favourable score grants nothing biologically resembling immunity; behavioural risks dominate most common diseases regardless of genotype, and families carrying rare high-impact mutations know better than population scores whom to watch. Conversely, an unfavourable finding can mislead both ways — overstated dread for modest relative effects, and appropriate vigilance missed when positive results hide inside un-gated reports that never reached counselling. This last channel is why regulators insist pathogenic-cancer variants undergo laboratory confirmation and professional interpretation before acting: the biology of one BRCA letter is settled science, while the psychology of receiving it unsupervised is a documented hazard. Consumers holding arrays should additionally remember that today’s blank output reflects yesterday’s questionnaire, not proved absence — a structural caveat no slick interface conveys honestly.