No-Calls in Raw DNA Data: What -- Really Means

Every raw file has positions marked --, 00 or ./. where the chip could not decide. Why they happen, what call rate is normal, and what a bad file looks like.

Scroll through any raw DNA file and you will find lines that end not in letters but in --. AncestryDNA writes them as 0 0; a VCF writes ./.. These are no-calls: positions where the chip took a reading but the software refused to assign a genotype. They are normal, they are informative in their own way, and the number of them is the quickest health check you can run on a file.

How a chip decides a genotype

A genotyping array does not read DNA letter by letter. For each SNP it carries probes that bind the DNA near that position, and a chemistry step lights up one colour for one allele and a different colour for the other. The scanner measures the two intensities. A homozygous sample lights up strongly in one colour, a heterozygous sample lights up in both.

Software then places each sample’s reading into one of three clusters - AA, AB, BB - by comparing it with the pattern seen across thousands of other samples. Each call gets a confidence score, and readings that fall between clusters, or that are too dim to place, score below the cutoff. Those become no-calls. The default threshold on Illumina arrays, which most consumer services use, is deliberately conservative: it would rather report nothing than report a wrong letter.

Why a particular SNP fails

Some no-calls are about you, some are about the SNP, and most are about the probe.

  • Probe chemistry. A few percent of probes on any array simply perform poorly - they bind weakly, cross-react with a similar sequence elsewhere in the genome, or sit in a region that is hard to amplify. These SNPs no-call in many people, and different services no-call different ones because their arrays differ.
  • Nearby variation. If you carry another variant right where the probe binds, the probe binds poorly and the signal drops. This is why some no-calls cluster in families and in particular ancestries.
  • Copy-number differences. A deleted or duplicated region confuses the two-colour model; the RHD gene that determines Rh blood type is a well-known example.
  • Sample quality. Low DNA yield or degraded DNA from a saliva sample dims every probe at once, pushing many borderline calls below threshold. This is the case where no-calls are telling you something about the sample rather than the SNP.
  • Difficult regions. The HLA region on chromosome 6, the pseudoautosomal parts of X, and repetitive stretches have elevated no-call rates for everyone.

What a normal call rate looks like

Call rate is the share of SNPs that received a genotype. A healthy consumer file typically lands between 98 and 99.7 percent. A few thousand no-calls in a file of 600,000 to 700,000 markers is unremarkable.

Two situations deserve attention:

  • A call rate below about 97 percent usually indicates a poor sample. Services often re-run these before releasing results, but files at the edge do get through. If yours is low, it may be worth asking the company to reprocess or re-collect; that is usually free within their quality policy.
  • A no-call at a specific marker you care about is a gap, not a result. It does not mean you carry the common allele, and it does not mean the marker is absent from the chip. It means there is no information at that position from this file. Another service’s file may have called it.

What no-calls do to downstream analysis

A trait lookup that depends on one SNP returns nothing for a no-call at that SNP - Genespiral lists these as markers it could not evaluate rather than guessing. Ancestry estimates and relative matching are barely affected by a normal number of no-calls, because they draw on hundreds of thousands of markers and the missing ones are scattered. Comparisons between two files should drop positions that are no-called in either, which is why a naive “how many lines differ” count between two of your own files exaggerates the disagreement.

The DNA file report on the Genespiral dashboard shows your call rate and how many of the trait markers it reads are present in your file, computed locally, so you can see where the gaps are before running an analysis.

Things that look like no-calls but are not

  • Single letters on X, Y and mitochondrial DNA in a male file are hemizygous calls, not partial no-calls. See our explainer on reading the two letters.
  • DD, II and DI are insertion/deletion genotypes, not error codes.
  • Missing lines. A marker that is not in the file at all was not on the chip. That is a coverage difference, not a no-call, and it is the more common reason a lookup finds nothing. Our comparison of which test gives the most useful raw data covers what each chip includes.

The takeaway

No-calls are the array being honest about uncertainty. A sprinkling of them is the sign of a system working as designed; a flood of them is a sample problem worth raising with the company; and a no-call at the one marker you wanted is simply a question the file cannot answer.

This article is educational only and is not medical advice.

References

Further reading