How Aimosti re-analyses your genome
You bring DNA data you already own, a genotyping-chip export or a whole-genome file. We run it through a fixed, versioned pipeline and show you what public databases say about your variants, with each claim attributed to its source. We never take a sample, sequence, test, or diagnose. Every clinically relevant match is reported the same way: this is a literature match, not a diagnosis. Confirm it with a clinician and a targeted, validated gene test before acting on it.
The pipeline, step by step
- Ingest & detect. We read your uploaded file, detect its format and genome build, and check which sample it contains.
- Normalise to GRCh38. Variants are left-aligned and normalised (and lifted over if needed) to one reference build, so every lookup is coordinate-exact.
- Annotate. We annotate against current public databases: consequence prediction plus clinical and population-frequency context.
- Restrict to curated panels and listed variants. Matching is bounded to reviewed panels rather than the whole genome. On a sequenced file the wider ClinVar screen reads only the exact variants ClinVar lists as disease-causing, in genes chosen from ClinGen's curations. Every result can be traced to a list that a named body maintains.
- Apply each module's rules. Clinical findings, carrier status, the wider ClinVar screen, pharmacogenomics, polygenic scores, traits and ancestry each restate sourced facts through their own lens.
- Freeze the report. The result is pinned to the exact source versions used, so your report never silently changes after it is issued.
Sources we restate
We build only on openly-licensed sources, and we surface each source's own claim and version. We do not invent our own classifications.
- ClinVar: clinical significance and review status (gold stars).
- gnomAD: population allele frequencies.
- VEP: variant consequence prediction.
- ACMG SF: the reportable secondary-findings gene list. Our guide covers which 84 genes it holds and why only those.
- CPIC: pharmacogenomic guidance.
- GWAS Catalog: trait associations, evidence-gated.
- PhyloTree: mitochondrial (maternal-line) haplogroups.
- ClinGen: gene–disease validity and adult actionability curations, which choose the genes of the wider ClinVar screen. Obtained from the Clinical Genome Resource (clinicalgenome.org), CC0 1.0, accessed 5 October 2026.
- Orphanet: the inheritance pattern of X-linked conditions, used to keep X-linked dominant conditions out of the wider carrier screen. Orphanet: an online rare disease and orphan drug database, © INSERM 1999, orpha.net, CC BY 4.0, accessed 5 October 2026.
- PGS Catalog: the published polygenic scores we restate, each named by its PGS Catalog identifier and the study it comes from.
- 1000 Genomes Project: European samples used to measure how linkage between a score's variants widens its reference spread, so a percentile is not read against a spread that is too narrow.
Absence of evidence
We never say "you don't have X." We say no matching variant was found in the regions your data covered, and we show that coverage. A gap in coverage is not the same as a clean result, and we never let it read that way.
When you supply a gVCF, we report per-gene callability: the share of each panel region covered at a read depth of at least 10, the common clinical-grade minimum for a confident germline genotype. A region below that depth is reported as not fully callable, so "no variant found" is never mistaken for "region confirmed clear".
Ancestry composition
We fit your genotypes against 31 reference components, 30 of them built from the Allen Ancient DNA Resource (open data, CC0), and report them in 17 regions. A region is a set of components whose reference populations are too close for a single file to tell apart reliably. The grouping is fixed in advance by how similar the components' allele frequencies are, not chosen per customer.
Each region gets an estimate and a range. The range is the spread of the estimate over 100 resamplings of your markers, widened at the top by the share we could not place, because that share may belong to any region.
We test the ranges two ways. On genomes simulated from the panel's own allele frequencies, a region's range held the true value 97 % of the time. That test is generous, because the simulated person comes from exactly the frequencies we fit against. On genomes stitched together from real reference people, whole chromosomes at a time, and fitted against a panel without them, the range held the known share 79 % of the time at 23andMe chip density and 67 % on whole genomes. Whole genomes do worse here: more markers make the range narrower, but the reference groups are no bigger, and that part of the uncertainty is not in the range. This second test errs low, because the reference people's own mixed ancestry counts as a miss.
Inside a region we print ranges only, rounded to 5 points: in simulation they held the true value 91 % of the time, which is why we don't print a single number for them.
These are measurements on simulated and reference genomes, not a promise about any one file. A file that fits the panel poorly says so on its card.
The Ashkenazi Jewish component is built from published allele frequencies. Its reference is 471 people from NCBI GEO GSE23636 (Bray et al. 2010, PNAS), submitted by Emory University (Bray, Warren). They were genotyped on a different array, and we use their allele frequencies only: no individual's genotypes are kept. Those frequencies cover 52 % of this panel's markers. On the rest, the component is modelled from its neighbours in the panel, so your file can tell it apart only on the covered markers. Part of an Ashkenazi Jewish ancestry will read as Western and Southern Europe, the region that holds its nearest reference, Italian–Balkan. It also works the other way: part of a Southern Italian or Balkan ancestry can read as Ashkenazi Jewish, which is why we name an Ashkenazi Jewish share only from 13 % up. On the map it is drawn where the reference people's recent ancestors lived, in Central and Eastern Europe. No other Jewish group has a component of its own: the Allen Ancient DNA Resource holds nine other present-day Jewish groups, with 5 to 10 people each, too few to build one from. Like every component, it measures how closely your genotypes resemble a reference group, and says nothing about religion or community.
Two questions, kept separate
We keep two questions separate: how sure is the field that this is real? (ClinVar review status, submitter count, replication) and how much would it matter if real? (effect size or penetrance). A large effect never masks weak evidence. See the glossary for the terms used in your report.
This page describes the method. The exact source versions used for any given report are pinned inside that report.