NEW YORK – Researchers have devised an approach that might make polygenic scores useful for people from mixed ancestral backgrounds.
Polygenic scores can be used to predict a person's likelihood of developing a trait or disease, but they are often not able to make such predictions outside of the ancestry group in which they were developed. This is a particular problem for people from recently admixed populations like African Americans, Afro Caribbeans, Central and Southern Americans, and Ethiopians and North Africans.
Researchers from the University of Tartu in Estonia developed a means of deconvoluting the different ancestries that an admixed individual may have to develop ancestry-specific partial polygenic scores that may then be used to generate a combined ancestry-specific polygenic score.
As they described in Nature Communications Thursday, the researchers applied this approach to examine various traits such as height and body mass index, to find that their combined scores improved trait predictability and could then enable such predictions to be applicable to a wider swath of the population.
"The latest developments of personalized medicine needed an extra step to be applied to individuals with more diverse origins, and here we tried to combine knowledge from homogeneous populations into a model that could work for recently admixed individuals," Davide Marneo from Tartu's Institute of Genomics, said in a statement.
The poor transferability of polygenic risk score can be modeled by applying a directional bias to the score, the researchers noted. They used this model to show that ancestry-specific partial polygenic risk scores and combined ancestry-specific polygenic risk scores could outperform traditional risk scores for an admixed population. They noted, though, that this relies on there being sufficiently predictive polygenic risk scores for the two ancestries that are combined in that admixed population.
In their approach, the researchers first teased out the different ancestry subsets within their genomes of interest. They then fed the non-admixed reference genomes for those ancestry subsets into their analyses. Along with summary statistics from genome-wide association studies, they calculated normalized partial polygenic risk score for each of those ancestry components as well as a traditional whole-genome polygenic risk score.
They compared population-wide polygenic scores and ancestry-specific partial polygenic scores for seven reference populations and three admixed populations — Egyptians, Ethiopians, and African Americans — for four traits. For many of the traits, there were significant differences in the score types across the three admixed populations.
Using samples from the UK Biobank that they divided into genetically African, East Asian, European, and admixed groups based on their principal components and Admixture algorithm scores, the researchers examined how well ancestry-specific partial polygenic scores could predict phenotypes. They found these partial scores could improve BMI and height predictions, the two traits they focused on in this cohort.
Combining those partial scores into a weighted-by-ancestry score further improved phenotype predictability, they reported. They noted, though, that the admixed individuals currently had to have some European ancestry for this prediction to work. Still, they noted that their approach might help extend the populations for which polygenic risk scores are applicable.
"Our work provides a solid proof of principle on the feasibility of using population genetic and molecular anthropology to boost the potential of personalized medicine," senior author Luca Pagani from both Tartu and the University of Padova in Italy added in a statement. "I hope our work can bring individuals of mixed ancestry one step closer to the benefits of personalized and predictive healthcare."