A new software method finds chromosome swaps that routine genome analysis misses

In Short. Software that keeps only the genuine two-way chromosome swaps among the thousands of false signals in standard genome data found 46 swaps in 16,131 people and gave a new diagnosis in five previously unexplained cases, though most of the swaps it found explained no illness.

Human DNA is packed into 23 pairs of long strands called chromosomes. Now and then two of them break and trade the broken ends. When no DNA is lost or gained, the result is a balanced translocation. Carriers are usually healthy unless a break cuts an important gene in two. A 2026 review puts these swaps at between 1 in 500 and 1 in 625 people.

The old way to see one is a karyotype, a picture of stained chromosomes under a microscope. The same review gives its practical resolution as about 5 to 10 million DNA letters, so smaller swaps escape it. Nor can it show which gene a break has hit. A common test for rare disease now reads the genome as a great many fragments, typically 150 letters long, and lines each up against a reference genome. This is short-read genome sequencing.

A fragment crossing a break lines up partly on one chromosome and partly on another, so sequencing should catch a swap. In practice the software that scans for such oddities flags thousands of possible breaks in each genome, and most are wrong, produced where repeated DNA makes fragments line up in the wrong place. These false structural-variant calls, or false positives, are why labs rarely report balanced translocations from sequencing. A 2019 review in Genome Biology put false-positive rates for short-read methods as high as 89%, depending on the type of change.

A team at Children's Hospital of Philadelphia (CHOP) led by Ramakrishnan Rajagopalan wrote software that keeps only flagged breaks that look like a genuine two-way trade. A real swap leaves a pair of joins at almost the same spot on each chromosome. The software groups flagged breaks into such pairs and checks the raw fragments around each break. If any point to a third chromosome, the call is dropped as a repeat artifact. Each surviving call comes with its breakpoint at single-letter resolution, the exact DNA letter where each chromosome broke.

They ran it on existing genome data from 16,131 people in five groups. It found 19 swaps among 3,202 presumed-healthy volunteers in the 1000 Genomes Project, 6 among 2,875 children with birth defects and their parents at CHOP, 9 among 3,574 patients in GREGoR, a rare-disease research consortium, 11 among 6,071 patients and relatives in the Undiagnosed Diseases Network, a National Institutes of Health program, and 1 among 409 newborns sequenced at CHOP.

In the birth-defects group all six swaps were confirmed by Sanger sequencing, an older method that reads a short stretch of DNA letter by letter. Six in 2,875 is one per 479 people, close to the long-cited estimate. GREGoR and Network calls were checked against the raw fragments instead. As a test of the method, it recovered all eight swaps that earlier testing across the patient groups had already found.

For patients the measure is diagnostic yield, the number of people who get an explanation for their condition. Of the 27 swaps found in the four patient groups, 8 explained a patient's condition, 3 already known and 5 new. In each new case the broken gene matched the patient's features. ENG causes hereditary hemorrhagic telangiectasia, a blood-vessel disorder, and that patient had nosebleeds and an abnormal artery-to-vein link in the lung. TANC2 causes a developmental disorder, though the authors call that patient's recorded features sparse. ARID1B, a cause of Coffin-Siris syndrome, a developmental disorder, was cut in a patient whose two earlier sequencing tests had been negative. EXT1 causes multiple exostoses, benign bony growths, and explained three affected relatives whose earlier gene panel, which included EXT1, found nothing. MED13L, linked to intellectual disability and heart defects, was cut in a child whose swap was already known but whose broken gene had not been found.

The work is a preprint and has not been peer reviewed. Most swaps were not diagnostic. Of the 27 in patients, 5 were candidates, 7 were of uncertain significance and 7 were carried by relatives without the condition. The software cannot resolve swaps whose breaks both sit in repeated DNA, and the authors cannot yet say how many real swaps it missed.

It shows that this one kind of chromosome change can be read from sequencing data many patients already have and traced to the gene it breaks. It does not replace a full genetic workup. In these groups it added two new diagnoses among 3,574 GREGoR patients, two among 2,027 patients in the Undiagnosed Diseases Network and one among 1,017 children with birth defects.