Missing single nucleotide polymorphisms in Genetic Risk Scores: A simulation study.

Area of Science:

• Population genetics
• Statistical genetics
• Genomic prediction

Background:

• Genetic risk scores (GRS) are used to predict phenotypes, but missing data on single nucleotide polymorphisms (SNPs) complicates their application.
• Common strategies for handling missing SNP data in GRS include imputation, omission, or using proxy SNPs.

Purpose of the Study:

• To evaluate the impact of SNP omission versus proxy SNP approaches on the estimation of effect sizes and predictive ability of GRS.
• To compare weighted and unweighted GRS with a small number of SNPs under varying missing data proportions (20-70%).

Main Methods:

• Simulated a dichotomous phenotype using real genotype data.
• Assessed the effects of omitting or replacing missing SNPs on GRS association with the phenotype, statistical power, and area under the receiver operating curve (AUC).

Main Results:

• SNP omission led to greater bias towards the null effect size compared to proxy SNP approaches.
• Omission resulted in reduced predictive ability and a more significant loss of statistical power.
• The predictive ability of weighted GRS is sensitive to the availability of high-weight SNPs.

Conclusions:

• Proxy SNP strategies offer advantages over omission for maintaining the accuracy and power of GRS in the presence of missing data.
• Careful consideration of SNP selection and missing data handling is crucial for effective GRS application in genetic prediction.