NEW YORK — How well a polygenic score developed in one population can be applied to another depends on the genetic distance between the groups, a new study has found.
Polygenic scores can be used to gauge a person's risk of developing disease by combining information on a range of genetic risk loci. But polygenic scores developed among people of one genetic ancestry are often less predictive among individuals of other genetic ancestries, raising concerns that such scores are mostly applicable to individuals of European ancestry where most data have been collected.
Using data from the UK Biobank, researchers led by Aarhus University's Bjarni Vilhjálmsson examined how portable 245 different polygenic scores were between different genetic ancestry groups living in the UK. As they reported in the American Journal of Human Genetics on Thursday, the researchers found that a polygenic score developed in a UK ancestry group was less predictive when applied to other European-origin ancestry groups and was even less predictive when applied to groups with ancestral origins outside Europe.
"Our analysis demonstrates a canonical relation between genetic distance and predictive performance for most phenotypes," Vilhjálmsson and colleagues wrote in their paper.
The UK Biobank includes genetic data on nearly half a million participants. While most are of British or European ancestry, the dataset includes tens of thousands of individuals of other genetic ancestry who live in the UK. By using this dataset to examine polygenic score portability, the researchers aimed to avoid the biases that may arise when using data from different studies that may have been generated in slightly different ways as well as to account for environmental differences.
Using principal components analysis, the researchers split their UK Biobank dataset into nine different ancestry groups, largely centered on geography. These include Northeast and Southern European groups as well as Middle Eastern, South Asian, East Asian, Caribbean, and Ashkenazi Jewish groups.
They then used the 391,124 individuals in their dataset of Northwestern European ancestry to train polygenic scores for 245 different phenotypes and tested how well those scores performed in the eight other ancestry groups.
Overall, they noted a low portability of polygenic scores between groups. In particular, their predictive performance fell to 93.8 percent in Northeast Europeans, 85.6 percent in South Europeans, and to 72.2 percent in the Middle Eastern group and 85.7 percent in the Ashkenazi Jewish group. The scores' performance was even lower among East Asian, Caribbean, and West African groups, at 48.6 percent, 25.2, and 18 percent, respectively. The performance decreased roughly linearly with increased principal component distance from the Northwest European ancestry training set.
"These analyses demonstrate that prediction already drops off within European ancestries and reduces globally in proportion to genetic distance," Vilhjálmsson and colleagues wrote.
The researchers added that these portability results were largely consistent across phenotypes — with some exceptions including hair color, tanning ability, and some blood measurements — which suggested to them that the predicted loss of accuracy could potentially be calculated.