A Bigger Set of Genomes Helps Genetic Studies Cover More of the World

Researchers at Kiel University in Germany have built a genetic reference library from the whole genomes of 490,319 UK Biobank participants. They report that it reconstructs the DNA a study did not sequence more accurately than the widely used TOPMed r3 panel, for most of the world's population groups but not for all of them.

Most genetic studies read only a sample of each person's DNA and work out the rest by matching it against a library of fully sequenced genomes, a step called imputation. The quality of that library sets the limit on what a study can find. The authors say they set out to give UK Biobank researchers an accurate option for diverse global populations. Their report appeared Oct. 30, 2026, in Nature Genetics, open access.

Lars Wienbrandt, David Ellinghaus and colleagues at the Institute of Clinical Molecular Biology measured the new panel against TOPMed r3, built from 133,597 genomes. Sorting variants onto the correct copy of each chromosome was more accurate with the new panel in 25 of 26 population groups from the 1000 Genomes Project, and in every broad continental group except the American one. Filling in missing genotypes improved for 4 of 5 continental groups and for 12 of the 26 smaller ones, while TOPMed r3 stayed better for the African group at every frequency range tested. East Asian samples gained the most, 24.39% on the paper's main quality measure. For rare variants in many populations, the authors write, accuracy remains low.

The team also reran published COVID-19 studies from Italy, Germany, Spain and Norway. In the Italian data, the new panel produced a statistically significant association at a known COVID-19 risk region where the TOPMed-based analysis had produced none, and it narrowed the list of variants likely to be responsible. The authors built the panel they are measuring.

The panel runs inside the UK Biobank Research Analysis Platform through a version of the group's EagleImp software, which is open source. Across their benchmark datasets, their runs took about half as long as the TOPMed imputation server.