Polygenic risk scores in diverse population

Question

[dandanli0365]

I begin to calculate polygenic risk scores. And i have had some problems. Gwas summary results were obtained from European samples, and they were used to calculate polygenic risk score in east asian samples. Results show that the performance is poor. There are many reasons for this, such as allele frequency, LD etc. Can someone know how to adjust SNP effect size or other ways for better performance in target samples?

[Kevin Blighe]

I don't believe there is anything to do this, currently. Obviously the ethnic groups should match.

You may consider either of the following:

• using just a subset of the SNPs that appear to be informative to East Asians
• use the full panel of SNPs but adjust the final derived scores based on population differences as measured by, e.g., PCA or IBS

Kevin

[dandanli0365]

thank you very much, could you please give me a paper about scores based on population differences?

Answer 1

[Yean]

1. Population consideration

The most appropriate method is to use summary statistics from a discovery population corresponding to the ethnicity of your target group

However, If they are not available, You may consider using either unweighted GRS (SNP's weights will be assigned equally) or weighted GRS with multi-ethnic summary statistics (SNP's weights will be averaged globally).

1. Model Performance

If you have whole-genome summary statistics, what about clumping it with different thresholds and evaluating the predictive ability of those candidate summary statistics. As a result, you will get the most optimal model with highest performance.

[dandanli0365]

Thank you and both methods are great. But i still feel a little confused. How can i get multi-ethic summary statistics? which website can download it?

[Yean]

you can explore the summary statistics from GWAS catalog or a public consortium such as cardiogram (coronary artery disease).