Polygenic Risk Scores That Predict Common Diseases Using Millions of Single Nucleotide Polymorphisms: Is More, Better?

Abstract

Following a decade of slow progress in improving the genetic prediction of common diseases with tempered expectations about the ultimate predictive ability of polygenic risk scores (PRSs),3 the emerging zeitgeist is that PRSs based on millions of single-nucleotide polymorphisms (SNPs) can identify individuals who are at increased risk for common diseases. In recent studies, this increased risk is claimed to be similar to conditions associated with known causal genetic mutations. Thus, the argument goes, it is time to start thinking about implementing PRSs in routine clinical care (1, 2). PRSs are generally calculated using SNPs that are statistically significantly associated in genome-wide association studies (GWAS). These SNPs are replicated across studies and more likely implicated, directly or by association, in the development of the disease. The predictive ability of PRSs improves when more SNPs are added, either by new GWAS “hits” or by lowering the statistical significance threshold. Several recent studies have used a new statistical method, LDpred, which does not select SNPs but optimizes the weights for all genotyped SNPs using their GWAS weights, the correlation between neighboring SNPs (known as linkage disequilibrium), and an estimate of the proportion of SNPs that are expected to contribute to risk and have nonzero weights (3). When this proportion is unknown, researchers may apply the method for several hypothetical proportions and select the one at which PRS explains the highest percentage of phenotypic variance. When LDpred is applied to millions of SNPs, it is unsurprising that most of them have negligible impact on disease risk. Researchers disagree whether the inclusion of millions of SNPs with close to zero effects in PRSs makes sense. From a bioinformatics perspective, there is no need to exclude them: SNPs with close to zero effects that don't improve PRSs won't make them worse either. …