Runs of homozygosity: what identical stretches of DNA record, and why Finns carry more

Two copies, one ancestor

Each of the 22 autosomes comes in two copies, one inherited from each parent. Over most of their length the two differ here and there, at the positions where a person is heterozygous. Inside a run of homozygosity they do not: position after position reads the same letter twice. A long run forms when the same stretch of an ancestor's chromosome reaches a person down both sides of the family tree, so that they inherit one ancestral piece twice^[1].

F(ROH) is only as comparable as its minimum. A study that counts runs from 500 kb adds up far more homozygous genome than one that starts at 3 Mb, so the figures in this guide always come with the length they were counted from.

Length works as a clock

Each meiosis cuts the chromosomes and splices them. A segment that travels from one ancestor down two lines of descent is cut along both, so the further back the ancestor lived, the shorter the piece that arrives intact on both copies. Thompson gives the rule: when two copies are separated by m meioses, the lengths of the segments they share follow an exponential distribution with a mean of 100/m centimorgans. The paper's worked example is a time depth of 25 generations, which is 50 meioses and an average of 2 cM^[3].

So runs of 3 Mb and up, the ones the report counts, mostly come from ancestors within the last few dozen generations, and runs under 1 Mb from ancestors much of a population shares. Kirin and colleagues put it in one sentence: offspring of cousin marriages have long runs, while the numerous shorter tracts relate to shared ancestry tens and hundreds of generations ago^[4].

The averages hide a wide spread, and that limits what one genome can say. McQuillan's paper gives the expected inbreeding coefficient of a first-cousin couple's child as 0.0625 with a standard deviation of 0.0243, because which segments pass down each line is a matter of chance; the groups it compared, sorted by pedigree, overlapped considerably in F(ROH)^[2]. A single genome's total is one draw from that spread.

Short runs, long runs and the histories behind them

Pemberton and colleagues mapped runs in 1,839 people from 64 populations with 577,489 markers and found three length classes. Short runs of tens of kilobases probably reflect ancient haplotypes. Intermediate runs of hundreds of kilobases to several megabases probably result from background relatedness in a population of limited size. Long runs of multiple megabases probably result from recent relatedness between a person's parents^[5]. Long runs were higher and more variable in most populations from the Middle East, Central and South Asia, Oceania and the Americas than in Africa, Europe and East Asia, and higher where consanguineous marriage is common^[5].

Short runs carry most of the total, even where long runs are frequent. Kirin's survey of worldwide populations found that in every one, including the most inbred, more of the genome sat in runs of 0.5 to 2 Mb than in runs over 2 Mb^[4]. A total built mostly from long runs points to recent shared ancestry; one built from many intermediate runs points to a small population, and the two can sit in the same genome.

A 2018 review of how runs are found in microarray and sequence data put the point in one line:

Why Finnish genomes carry more

Finland's inland north and east were settled late, from the 1500s^[8], and the youngest sub-isolates in the north-east were founded only 300 to 400 years ago. Jakkula and colleagues read the genetic pattern as multiple bottlenecks from consecutive founder effects^[6]: each settlement drew a small sample of genomes out of an older population, and its descendants then married mostly among themselves. The Finnish DNA guide traces the same history through the Y chromosome and the east–west divide.

They measured it in 1,395 people, from ten Finnish subpopulations plus comparison samples from Helsinki and Sweden, counting runs longer than 1 Mb from 231,116 markers. Mean homozygosity rose from 0.9 percent of the genome in the early-settled south coast to 2.0 percent in the North Kainuu isolate. The number and length of runs were highest in the youngest subpopulations, where up to 90 percent of people had at least one run longer than 5 Mb; in a US sample, 9.5 percent did^[6].

Martin and colleagues saw the same geography in segments shared between people rather than within one genome, which is the same inheritance seen across two people instead of two copies. In 43,254 Finns, people around Kuusamo, a late-settled area in the north-east, shared on average about 60 Mb in segments longer than 3 cM with others born nearby; around Helsinki, Turku and Tampere the figure was about 5 to 15 Mb^[9]. Across countries, two unrelated Finns shared on average 107.0 cM, two Swedes 22.9 cM^[9].

Finland is not the extreme case. In high-coverage whole genomes analysed in 2026, the Finnish samples of the 1000 Genomes Project averaged about 63.9 Mb in runs longer than 1 Mb, and 40 people from the Faroe Islands about 82.5 Mb, more than any group in the study; the Faroese also carried more runs in the 5 to 15 Mb range, which the authors read as a more recent or stronger bottleneck^[7]. On chip data, endogamous island communities in Dalmatia and Orkney carried several times more than cosmopolitan samples^[2].

The link to the Finnish Disease Heritage

A recessive condition appears when both copies of a gene carry a disease-causing change. Inside a run, both copies are the same ancestral copy, so a variant that one founder carried can arrive twice. Martin and colleagues note that long runs are enriched for deleterious variation, and they found that two Finns who carry the same variant from the Finnish disease database share a stretch of genome around it about 30 percent of the time or more, an order of magnitude above the background rate^[9].

The same settlement history drew both maps. FinDis records that for most heritage diseases the birthplaces of patients' families concentrate in the late-settlement area populated from the 1500s, and that Northern epilepsy is found in Kainuu, near the eastern border^[8], where Jakkula measured the highest homozygosity^[6]. The Finnish Disease Heritage guide covers the 39 diseases themselves, and carrier status vs your own risk what one copy of a recessive variant means. A run says nothing about which variants sit inside it, so this module reports no carrier results; the carrier panel reads them directly.

What each kind of file can measure

Most of what is known about runs comes from chips. McQuillan used a 300,000-marker chip, Jakkula 231,116 markers and Pemberton 577,489^{[2, 6, 5]}. The consumer chip exports we measured hold between 563,320 and 955,958 rows^[10], so a chip of that size can, in research hands, find runs of a megabase and longer. Density sets the floor: with a panel of 3 million markers, runs as short as 100 kb can be detected reliably, and Han Chinese genomes that showed about 130 Mb of runs on a panel of about 400,000 markers showed about 510 Mb on the dense one^[4]. A typical genome differs from the reference at 4.1 to 5.0 million sites^[11], and a whole-genome VCF lists them all: by our arithmetic, more than a thousand per megabase on average.

Our report does not measure runs on chips. Its scan counts heterozygous calls in each megabase of a whole-genome call set, and it was built and checked on that kind of file only. The report says so in place of a number. Exomes, targeted panels and low-pass or imputed genomes have the same problem in another shape, and the report refuses them too, naming the evidence it found.

What the report shows

In the report the module follows the ancestry section, under the heading Runs of homozygosity. The scan splits the autosomes into 1 Mb bins, treats a bin as homozygous when it holds at least 50 variant calls of which at most 2 are heterozygous, joins neighbouring bins, and keeps runs of 3 Mb or more. A bin with too few calls counts as unmeasured, never as homozygous. The total, divided by the 2,875 Mb of autosome the scan counts, is the share the report gives. The Finnish DNA guide draws the scan on an invented example.

The card leads with a band headline and three figures: the share of the autosomes in long homozygous stretches, to one decimal; the number of stretches of 3 Mb or more; and the longest stretch. Below them come the band's description, a short account of what runs mean and a caveat. The report's summary list shows the headline alone, never the percentage. A refused file gets the headline Not measured for this file and the reason.

The band edges are round numbers chosen for reading, not clinical thresholds. They also sit on a different scale from the published Finnish figures: Jakkula counted runs from 1 Mb, the report from 3 Mb in whole 1 Mb bins, so the same genome usually shows a smaller share in the report than it would in that study^[6]. The caveat on every card names descent from a founder population, marriage within a small community and recent shared ancestry together, and says the measurement does not distinguish between them.