What Is Genotype Imputation (and Can You Trust It)?
Imputation fills in genotypes a chip never measured by matching you against sequenced reference genomes - useful, cheap, and inferred rather than directly read.
Your DNA chip reads a few hundred thousand positions, yet ancestry and research tools often seem to know about many more. The trick that bridges that gap is called imputation, and it is worth understanding both its power and its limits before you lean on it.
What imputation does
A genotyping chip measures a fixed, pre-selected list of positions directly. Imputation statistically fills in genotypes the chip did not measure - it infers the letters at unmeasured positions rather than reading them off your sample. It does this by matching your measured variants against a large reference panel of fully sequenced genomes, and using the patterns it finds there to make educated calls about the gaps.
The result is a much denser picture of your genome, built cheaply on top of inexpensive chip data. That is why imputation is so widely used: it stretches a modest set of measurements into something far larger without the cost of sequencing.
How it works, intuitively
DNA is inherited in chunks, not shuffled letter by letter. Nearby variants tend to travel together in blocks that recur across a population, a pattern known as linkage. Imputation exploits this. If you carry a particular combination of measured variants, and thousands of sequenced people in the reference panel who share that combination also share a specific letter at an unmeasured position, then you very likely carry that letter too.
So imputation is not guessing in the dark. It is pattern-matching against real, fully sequenced genomes and asking: given what we measured directly in you, what almost certainly sits in between?
Why it is inference, not measurement
Here is the crucial distinction. An imputed genotype is a probability, not a direct read. The chip physically interrogated the positions it measured; imputation calculates the likely value everywhere else. Most imputed calls come with a confidence score, and the high-confidence ones are usually right - but “usually right” is a statistical statement, not the certainty of a direct measurement.
This matters most when you drill down to a single position. If a specific variant is important to you, it is worth knowing whether it was directly genotyped or imputed, because the two carry different weight.
Where accuracy drops
Imputation is not uniformly reliable. Its accuracy depends on how well your DNA is represented in the reference panel:
- Rare variants are harder to impute. There are fewer examples in the panel to learn from, so uncommon letters are more often missed or miscalled.
- Under-represented populations get weaker results. Reference panels have historically been built mostly from a few well-studied populations, so people whose ancestry is thinly represented receive less accurate imputation.
- Unusual regions of the genome, where inheritance patterns are complex, are also more error-prone.
None of this makes imputation useless. It means the confidence you place in an imputed call should scale with how common the variant is and how well your background is covered.
What is in your downloadable file
A helpful practical point: the raw file you download from a consumer service usually contains the directly genotyped positions - the ones the chip actually measured - rather than the full imputed set. Services may use imputation internally to power ancestry and reports, but your export typically reflects direct reads. That is a good thing, because it means the letters in your file are measurements you can trust as measurements.
To see how those directly measured positions are structured, our tour of SNP chips versus whole genome sequencing explains exactly what a chip captures and what it leaves out. If you want the fullest possible view of a specific variant, sequencing reads it directly rather than inferring it.
The bottom line
Treat imputation as a smart, well-founded estimate that expands your data enormously for very little cost - excellent for the big picture, and best sanity-checked for any single rare or personally important variant. When you explore your own file with on-device DNA analysis, you are working with your directly measured genotypes, kept private in your browser, which is the most solid ground to stand on.
This article is educational and is not medical advice.