Real findings inside a UI built for anxiety. Solid clinical-grade variants are right next to GWAS-derived noise, undifferentiated. The product would be a 4 if they fixed two things — but they haven't.
What SelfDecode is.
SelfDecode is a direct-to-consumer platform that takes a raw genetic file (typically from 23andMe or AncestryDNA), or runs its own genotyping kit, and produces personalised reports across health, traits, supplements, diet, and longevity. The product layers polygenic risk scores, single-variant calls, and AI-generated recommendations on top of your raw genotype.
The pitch is "your DNA, decoded." For a small subset of well-validated variants — ApoE for Alzheimer's risk, Lp(a) loci for atherosclerotic cardiovascular disease, MTHFR for one-carbon metabolism, BRCA family for cancer risk — that pitch is largely true. For the rest of the recommendation surface, it's not.
What actually worked.
Credit where it's due: SelfDecode correctly flagged my Lp(a) genetic risk profile. When I subsequently got a clinical Lp(a) measurement at YEARS, it came back at 256 nmol/L — well into the high-risk zone, exactly as SelfDecode's polygenic prediction had warned. Same for ApoE — the genotype call matched a clinical genotyping result.
That's a real product win. For variants with strong GWAS evidence and well-characterised clinical phenotypes, SelfDecode's calls are accurate enough to take seriously and follow up on with a clinician. If you have not yet done genetic testing and you specifically want to know your Lp(a) genotype, your ApoE status, your MTHFR, or your BRCA-family risk in advance of clinical testing, the platform delivers that.
Where it falls apart.
Every other surface of the product is the issue.
Information overflow ten times over. A SelfDecode dashboard returns hundreds of "personalised recommendations" — prioritise X foods, avoid Y, supplement Z, sleep this many hours, eat at this time of day. With no apparent prioritisation by evidence weight or by your specific situation. The default state of using the product is anxiety: every report opens with two dozen yellow flags and no clear "do these three things first."
Pseudoscience next to evidence-based stuff, indistinguishable. This is the bigger structural problem. A recommendation from a 2018 RCT in 8,000 humans on ApoE-modulated dietary fat is rendered in the same UI, with the same visual weight, as a recommendation derived from a 12-mouse study on a niche metabolite. The user has no way to tell from the UI which one to trust. SelfDecode doesn't surface evidence quality (Oxford CEBM levels, GRADE, anything) at the recommendation level. Everything looks equally authoritative.
"Solid evidence and animal-model speculation rendered with the same confidence — the UI has no way to tell them apart."
Recommendations are mostly generic. Even when the genotype is correctly called, the actionable advice frequently boils down to "eat real food, exercise more, manage stress, sleep more." Things you'd already do. The genotype-specific personalisation that justifies the subscription is rarely present at the recommendation layer — it's mostly at the report level.
Anxiety-inducing UX by default. The platform reads like a medical worry list rather than a tool. Reports surface every conceivable elevated risk in your genotype, regardless of effect size or actionability. For users who are pre-disposed to health anxiety, this is the wrong product. The "you have variant X, which has been associated with Y" framing is technically accurate and clinically irresponsible if it terminates without "and here is what (if anything) to do about it."
Pros
- Correctly flagged Lp(a) genetic risk — confirmed against clinical lab values
- ApoE and other well-validated single-variant calls are reliable
- Genotype upload supports raw 23andMe / AncestryDNA files — no extra kit needed
- Periodic updates as new GWAS evidence is published
Cons
- Information overflow — hundreds of recommendations, no prioritisation
- Pseudoscience and RCT-grade evidence rendered indistinguishably
- Most recommendations are generic and would apply to any user
- Anxiety-inducing UX, especially for non-clinicians reading raw risk reports
- Polygenic risk scores feel weightier than they are — most have small effect sizes
Who it's for.
SelfDecode is the right product if you specifically want to look up well-validated single-variant calls (ApoE, Lp(a), MTHFR, BRCA) before paying for clinical genotyping, and you can mentally filter out the noise. It's the wrong product if you want a personalised health platform — the recommendations layer is too unfiltered to act on responsibly.
For most users, a single visit to a clinician with a 23andMe raw file in hand will produce a more actionable conversation than a SelfDecode subscription. For Lp(a) specifically, just get a clinical-grade direct measurement (it's ~€40 in Germany, similar in the US through Function Health or LabCorp).
Verdict · 2 / 5
The genotype layer is real — that's the score floor. Everything around it is the score ceiling. Until SelfDecode separates evidence-grade recommendations from animal-model curiosities in the UI, and surfaces a clear prioritisation, it produces more anxiety than insight. Skip unless you specifically want a single-variant lookup tool.
Related reviews.
Reviewed by Niko Hems · Last updated