Textbook genetic principles come to life when we have the opportunity to scrutinize our own data. We learn that half of our autosomal DNA comes from our father and half from our mother, and then we see it graphically illustrated, as Family Tree DNA’s Chromosome Browser shows for chromosome 11:

Figure 1. Chromosome comparison showing where a tester matches his mother (orange) and father (blue). The comparison was done using the chromosome browser tool at familytreedna.com.

The microarrays (“chips”) currently used by the genetic genealogy companies test about 600,000 to 700,000 single nucleotide polymorphisms (SNPs; positions known to vary) scattered over the whole genome. Each of these SNPs has two possible versions (alleles), and the probes on the chip will hunt for the presence of each allele.1 Every allele found in the child can be found in the mother (the band color coded orange) and every allele found in the child can be found in the father (the band color coded blue). Each chromosome is one continuous segment.

This level of genetic literacy lets us spot anomalies. Sometimes (rather frequently, as it turns out) there are gaps. Figure 2, clipped from a GEDmatch comparison of a parent and child, shows a break in the blue band. The yellow and green lines show SNPs where the child matches one or both of the parent’s alleles, while the red lines near the center show a cluster of SNPs where a child does NOT match his parent. This leaves a gap of almost 50,000 bases.

Figure 2. Chromosome comparison showing a mismatch between child and parent. The comparison was done in the one-to-one tool at GEDmatch.com with the zoom level set to 5,000 pixels.

What is the explanation for this gap? Leaving aside the facetious suggestion that some alien DNA has infiltrated the chromosome, could the child have a bunch of mutations clustered in this location? That’s exceedingly unlikely, for the mutation rate for autosomal SNPs is very low, on the order of one or two changes per 100,000,000 bases.

One possible explanation is that it is due to a limitation in the testing technology, albeit one with some interesting implications. The gap may be a microdeletion (Conrad, 2006).  Microdeletions are generally defined as a loss of 1,000 to 5,000,000 bases, too small to see under a microscope with ordinary staining techniques. Recall that the chip technology looks for the presence of an allele. The genetic genealogy companies do not quantify the amount of an allele. If the base calling software sees both an A and a G for a particular SNP, it will report a heterozygous genotype of AG. If it sees only A, it will report a homozygous genotype of AA, or if it sees only G, it will report a homozygous genotype of GG.

With a deletion, one chromosome in the child will actually be missing any result for a SNP in the vicinity, and the allele from the other chromosome will be reported as homozygous. This leads to some contradictory findings: the child may be AA and the father may be GG, a “Mendelian inconsistency.” According to the principles first discovered by Gregor Mendel, the father of modern genetics, the child should have at least one G.

Figure 3 shows how the actual and reported genotypes might differ in a case where the child does not match his father for a SNP. The missing allele is denoted with an x. In actuality, the child is neither homozygous nor heterozygous: he is hemizygous. It is not possible to tell without more information whether the deletion was also present in the parent (inherited) or appeared for the first time in the child (de novo).

Figure 3. Hypothetical example showing how a missing allele can affect reported genotypes.

The impetus for this column came from numerous questions about these mysterious gaps, posted on various genetic genealogy forums and mailing lists over the years.  It’s difficult to estimate the frequency this way. Did these queries arise from oddities and outliers, or were they perhaps the tip of the iceberg, surfacing a common phenomenon?

To approach this question, I solicited GEDmatch IDs for parent/child kits. I informed the participants that I planned to write a column, but I did not reveal the nature of my request in order to avoid ascertainment bias, where respondents might be more likely to send just the “interesting” cases.

The results were indeed intriguing. Out of a total of 86 parent/child combinations, only 11 (13%) displayed the expected number of 22 segments (one for each chromosome). The overall average was 24.7 segments, with gaps of varying sizes as shown in Figure 4.

Figure 4. Distribution of 251 mismatch (gap) sizes in 86 parent/child comparisons at Gedmatch.com.

That rate was much higher than I expected a priori (that’s why we collect actual data instead of merely theorizing).  It certainly convinced me that the topic merited a column, but it led inexorably to another question. How can we tell if these gaps are actually microdeletions, or if they are due to some other limitation in the testing process?

Referring back to Figure 2, there is an isolated red line toward the right edge, which does not generate a gap. There is a mismatch, but it is surrounded by matching SNPs. The genotyping process is not perfect, and occasional miscalls are bound to occur. GEDmatch and the testing companies tolerate a mismatch or so before declaring an end to a segment, provided it is embedded in a long continuous run of matching SNPs.  “Long” is not an absolute quantity, and some algorithms may be more strict than others.

The mission of GEDmatch is to identify matching segments, not to analyze gaps. We are looking for mismatches that are clustered close together as a solid demonstration of microdeletions. David Pike has a suite of utilities for examining raw data, which will prove useful for digging deeper into the gaps. (This section is for those who like to get their hands dirty; others may feel free to skip ahead to the next section.)

One tool is called “Search for Discordant SNPs in Parent/Child in Raw Data Files.”  Discordant is synonymous with Mendelian inconsistency in this context. Figure 5 is a screen capture of some output from this utility, with columns for chromosome, reference SNP ID (rsid), position, and genotypes for the parent and child. Widely separated mismatches are included, but just eyeballing the results, it is clear that six closely spaced mismatches are found at about 112 megabases on chromosome 8.

Figure 5. Mismatched SNPs between a parent and child. Note the cluster of mismatches on chromosome 8. The comparison was done at http://www.math.mun.ca/~dapike/FF23utils/pair-discord.php.

Supplement 1 contains a spreadsheet for automating the calculations. When data from Pike’s output screen is pasted in to it, it produces summary information about the gap. 
Figure 6. Summary information about the gap on chromosome 8 (Figure 5). A spreadsheet for automating these calculations is in Supplement 1.

Another level of confirmation examines the gap SNP by SNP. Referring back to Figure 3, the discrepancy is detected because the child received an A from the mother. If the mother happened to contribute a G (being homozygous GG or heterozygous AG), then the child’s genotype would pass muster, masking the presence of a deletion.  Figure 7 shows all of the SNPs in the gap, using David Pike’s utility “Inspect a Shared DNA Segment in Two Raw Data Files.” 

Figure 7. All of the SNPs in the gap on chromosome 8 (Figure 5). These data were generated by http://www.math.mun.ca/~dapike/FF23utils/pair-discord.php

All the SNPs within the identified gap (and indeed for some distance beyond, not shown) are homozygous. A heterozygous result would have ruled out a microdeletion.