Why Chromosome X Needs Special Treatment
Many genotype-imputation procedures are described using diploid genotypes: each person has two allele copies at an autosomal site. Chromosome X does not follow that assumption uniformly. Under the usual XX/XY model, individuals with two X chromosomes are diploid across X, while individuals with one X chromosome have one copy through most of the chromosome. The pseudoautosomal regions (PARs), which have corresponding regions on X and Y, are an important exception: genotypes there are generally treated as diploid in both groups.
These differences affect how observed genotypes are represented, how imputation software calculates probabilities, and how dosages and quality checks should be interpreted. A value of 1 can mean a heterozygous diploid genotype or one alternate allele in a haploid genotype; without the sample’s ploidy and the variant’s region, the number alone is ambiguous.
The explanations below use the conventional XX/XY model to make the coding principles clear. Chromosomal complement is not always captured by a binary sex field: aneuploidy, mosaicism, and other biological variation can alter ploidy. Sample metadata should therefore be interpreted thoughtfully rather than treated as an infallible statement about every locus.
Regions, Ploidy, and Dosage
The PAR boundaries depend on the genome build. A variant’s chromosome coordinate must be interpreted against the same build used for the data and reference panel; a coordinate mislabelled by build can lead to the wrong ploidy treatment. The earlier material on genome build matching and allele harmonization is directly relevant: verify both the location and the REF/ALT allele alignment before interpreting dosage.
For a diploid genotype, the possible alternate-allele counts are 0, 1, and 2, corresponding to REF/REF, REF/ALT, and ALT/ALT. A haploid non-PAR genotype has only 0 or 1 alternate copies. Imputation returns a dosage that can be fractional because it is an expectation, not a literal fractional chromosome. For example, a haploid dosage of 0.8 represents uncertainty between the two possible haploid alleles, not a genotype containing 0.8 of an allele.
Some software accepts haploid non-PAR genotypes directly; other workflows use a diploid-equivalent convention for XY samples, representing haploid REF as 0/0 and haploid ALT as 1/1. Under that convention the stored dosage is 0 or 2, twice the haploid allele count. These conventions are not interchangeable unless conversion is deliberate and consistent. The correct interpretation must be established from the file format, software settings, and documentation.
Worked Example: Classifying Genotypes by Region and Ploidy
Consider three hypothetical samples: sample A has an XX complement, and samples B and C have an XY complement. At a PAR variant, their recorded diploid genotypes are A: REF/ALT, B: REF/REF, and C: ALT/ALT. At a non-PAR X variant, the calls are A: REF/ALT, B: REF, and C: ALT.
PAR variant. All three calls are diploid under the stated convention. The alternate-allele dosages are 1, 0, and 2, respectively. The total is \(1+0+2=3\) alternate copies among \(3\times2=6\) allele copies.
Non-PAR variant. Sample A remains diploid, so REF/ALT has dosage 1. Samples B and C are haploid, so REF has dosage 0 and ALT has dosage 1. The total is \(1+0+1=2\) alternate copies among \(2+1+1=4\) chromosome copies.
If the same non-PAR calls were stored using the diploid-equivalent convention, B’s REF and C’s ALT would be written as 0/0 and 1/1, with dosages 0 and 2. That representation does not mean those samples have two X copies; it is a software coding convention for haploid calls.
Allele Frequencies and Imputed Dosages
For a mixed-ploidy sample, a pooled allele frequency should count allele copies in the numerator and chromosome copies in the denominator. In the standard XX/XY model at non-PAR X, let \(d_i\) be the ALT dosage for sample \(i\), and let \(c_i\) be that sample’s number of X copies at the locus. The estimated ALT frequency is the total observed or expected ALT copies divided by the total number of X copies. The copy number is 2 for XX and 1 for XY under this model.
This formula uses dosage on the actual ploidy scale: 0–2 for diploid samples and 0–1 for haploid samples. If a pipeline instead stores XY non-PAR dosages on a doubled, diploid-equivalent scale, those values must be converted back to haploid counts before applying this formula. At a PAR site treated as diploid in all samples, the denominator is twice the number of nonmissing samples.
Worked Example: Pooled Non-PAR Allele Frequency
Suppose a hypothetical non-PAR variant has dosage 1.4 in each of two XX samples and dosage 0.7 and 0.2 in two XY samples. Assume these dosages were produced on the actual ploidy scale, with XX dosage in 0–2 and XY dosage in 0–1, and that all four samples have usable calls.
The expected ALT-copy total is \(1.4+1.4+0.7+0.2=3.7\). The total number of X copies is \(2+2+1+1=6\). Therefore,
The estimated ALT frequency is 0.6167, or about 61.7% of X chromosome copies in these samples. Dividing by \(2\times4=8\) would be wrong because it would count two X copies for each XY sample. If the XY dosages had instead been stored on a doubled scale, the values 0.7 and 0.2 would need conversion to 0.35 and 0.10 before the calculation.
Assumptions and Quality Checks
Ploidy-aware imputation assumes that the input genotypes, sample metadata, reference panel, and software settings agree about which samples are haploid or diploid at each position. It also assumes that PAR boundaries and chromosome coordinates correspond to the chosen genome build, that allele labels are harmonized, and that missingness or genotype uncertainty is represented in a way the software can use. These are data and model assumptions to inspect, not details that can safely be inferred from a dosage file alone.
Check the genome build and determine whether each X variant lies in a PAR or non-PAR region. Use the corresponding Y-region information where the analysis requires it.
Determine which samples are expected to be diploid or haploid at each region. Investigate discrepancies rather than silently changing them to fit a binary label.
Establish whether non-PAR XY calls and dosages use the haploid 0–1 scale or the diploid-equivalent 0–2 scale. Make any required conversions explicit.
Examine missingness, genotype classes, dosage ranges, allele frequencies, and unexpected heterozygosity separately by region and sample group.
At non-PAR loci, a heterozygous genotype in a sample expected to be haploid is a warning sign under the conventional model. It may reflect incorrect ploidy coding, a PAR-boundary or build error, a sample or metadata problem, or a biological exception. It should prompt investigation rather than automatic deletion or forced recoding. Conversely, apparent homozygous diploid calls in a haploid sample may be a valid diploid-equivalent representation; inspect the convention before flagging them.
Worked Example: Diagnosing a Non-PAR Coding Warning
A hypothetical non-PAR variant has these calls in four samples: two expected XX samples are REF/ALT and REF/REF; two expected XY samples are REF/ALT and ALT/ALT. The file header says that genotypes are diploid, but it does not document a haploid convention.
The XX calls are possible diploid genotypes. Under the usual model, however, an XY sample should have one allele at a non-PAR locus. Both XY calls appear heterozygous or diploid homozygous, which is inconsistent with a straightforward haploid call representation. The ALT/ALT call might be a valid diploid-equivalent encoding of haploid ALT, but the REF/ALT call cannot encode a single haploid allele under that same convention.
The appropriate next checks are the variant’s build-specific position, the input and imputation software’s ploidy settings, and the meaning of the genotype field. If the site is actually in a PAR, the expected ploidy changes. If it is truly non-PAR and the XY call remains heterozygous after the encoding is understood, investigate sample identity, metadata, variant harmonization, and biological exceptions. This small example does not establish that any particular sample is erroneous; it shows why an unexplained heterozygous non-PAR call is a diagnostic signal.
Why the Choice of Coding Matters
Ploidy errors can distort imputation inputs and downstream summaries. Treating a haploid call as an ordinary diploid heterozygote creates a genotype state that is not expected under the standard non-PAR model. Treating a PAR call as haploid discards a valid two-copy genotype. A denominator that assumes two X copies for every sample misstates the allele frequency, while inconsistent dosage scales can make values appear out of range or double-count alternate copies.
There are two broad approaches. A ploidy-aware approach represents non-PAR XY genotypes as haploid and PAR genotypes as diploid, using software that supports those states. A diploid-equivalent approach encodes haploid calls as homozygous diploid calls, typically 0/0 or 1/1. The latter can be convenient for tools that require diploid fields, but it must not be mistaken for observed diploidy, and dosage and frequency calculations must follow the chosen scale. Neither convention fixes incorrect sample metadata or boundary definitions.
After imputation, quality metrics and dosage ranges should be interpreted in light of the ploidy representation. A dosage above 1 can be valid for a diploid XX non-PAR sample, but not on a true haploid 0–1 scale. A doubled dosage for an XY sample may be valid if the pipeline explicitly reports diploid-equivalent values. The earlier material on imputation quality metrics and INFO scores provides general quality concepts; here the essential extra check is that the dosage scale matches the sample’s ploidy and genomic region.
No single X-specific check is sufficient. A reasonable review compares PAR and non-PAR behavior, checks male non-PAR heterozygosity under the actual encoding, verifies dosage ranges and missingness by sample group, and confirms that allele-frequency denominators count the right number of chromosome copies. Large group differences may indicate a technical or metadata issue, but they can also arise from genuine frequency differences; a diagnostic pattern is a reason to investigate, not a causal explanation.
Common Mistake
A second frequent error is to treat a software’s diploid-equivalent 0/0 or 1/1 representation as evidence that the sample has two X copies. The call may instead encode one haploid allele. Always distinguish biological ploidy from the software’s storage format.
Check Your Understanding
Answer the questions, then compare your reasoning with the solutions.
- At a non-PAR X variant, an XX sample has dosage 1.2 and an XY sample has haploid-scale dosage 0.6. What is their combined ALT-copy total and allele frequency?
- Why is a diploid-equivalent XY dosage of 2 not necessarily evidence of two X chromosome copies?
- A call is heterozygous in an XY sample at a site labelled non-PAR. Name two checks to make before deciding what to do with it.
- What denominator should be used for a PAR-site allele-frequency estimate if all \(n\) samples have diploid calls?
- Why must a chromosome build be checked when assigning X variants to PAR or non-PAR regions?
Solutions. 1. The total is \(1.2+0.6=1.8\) ALT copies. The two samples contribute \(2+1=3\) X copies, so the frequency is \(1.8/3=0.600\). 2. A diploid-equivalent value of 2 may encode one haploid ALT allele as 1/1; it is a file convention, not a claim about biological ploidy. 3. Check the variant’s build-specific position and the genotype encoding or ploidy settings; also review sample metadata and possible biological exceptions. 4. Use \(2n\) allele copies, assuming all calls are nonmissing. 5. PAR boundaries are coordinate-specific and build-dependent, so a mismatched build can assign the wrong ploidy model to a variant.