What is imputation? Why free tools can disagree
Your raw file contains genotypes for ~600,000–700,000 selected SNP positions. The genome has ~3 billion base pairs and millions of known variant sites — the chip only measures a small sample. Imputation is the statistical method that fills in the gaps — and that is why different tools can give different answers for the same SNP.
What imputation is
Consumer chips measure a sample of SNPs, not the whole genome. Imputation uses reference panels (e.g. 1000 Genomes, HRC) with known haplotype patterns: if your chip shows certain genotypes around a position, statistics can predict the unmeasured position with a probability. The result is not a measurement — it is a calibrated guess with an associated uncertainty.
Why tools give different answers
- Different reference panels: panel size and population mix affect accuracy, especially for non-European backgrounds.
- Different algorithms: Minimac, Beagle, Impute2 and others weigh linkage patterns differently.
- Uncertainty thresholds: one service may show a low-confidence guess, another filters it out.
- Rare variants: rare alleles impute worse than common ones — tools draw the line differently.
Same person, same raw file, two imputation tools → the same position can therefore get a different genotype. Neither of them "measured" it.
DNAHelper's position: show what was measured
DNAHelper does not impute. The raw file's genotypes are shown as they are — every position in your results is an actual measurement from your chip, not a statistical interpolation. If a SNP is missing from the file it is shown as missing instead of being replaced by a guess. That is a deliberate choice: for genetic questions "we do not know" is often more honest than "probably X".
When imputation is legitimate
Imputation is an established, valuable tool in genetics research — GWAS studies use it to reach millions of positions. The point is not that imputation is wrong, but that it is a different kind of data: a model statement, not a measurement. Qualified research tools report the uncertainty (info score, r²); consumer tools rarely make it equally clear.
Also: What is a SNP? · Why results differ · What happens to your file? · English overview