Reply to: Genome-wide association studies of polygenic risk score-derived phenotypes may lead to inflated false positive rates

Abstract

In their Matters Arising article, Uffelmann et al. 1 present simulation experiments suggesting that the approach used in 2 , where genome-wide association studies are based on polygenic risk score-derived phenotypes, may lead to inflated positive rates. We acknowledge the importance of confirming the increased number of false positive results by performing simulation experiments but have reservations about how these impact our interpretation of data and subsequent conclusions derived from the results in the original study.Our study determined each individual's polygenic risk score (PRS) for AD in the UK Biobank dataset. Using individuals within the extreme risk distribution, we performed a GWAS based solely on known genetic risk for disease, independently of diagnoses. Given that we were separating the two study groups by their genetics and not a well-defined trait or phenotype, we observed an excess of genome-wide significant associations (p < 5 × 10 -08 ), shown by a genomic inflation factor of 2.442. Since we expected this to occur (given the study design), we did not focus on these variants in our discussion and do not claim that the 246 independent loci are significantly associated with AD. Consequently, we do not consider this an increase in the number of loci associated with the disease. Instead, we adopted a more stringent threshold (p < 1 × 10 -15 ) to highlight highly significant variants associated with the published PRS extremes and potentially avoid false positives that were expected due to, for example, LD with our initial risk variants in the PRS calculation.Our view is that we should not use these results as those from a typical GWAS since p-values will generally be inflated/deflated for variants that were/were not associated in the base GWAS. Additionally, the effects will not translate to the typical case/control paradigm since variant frequencies will not necessarily reflect those in the general (control) population. We carefully considered these caveats before performing the study, and the results matched our expectations.A critical aspect of this work is that the study design we adopted is a step forward in allowing us the possibility of auditing previously established associations. For example, despite using summary statistics from 3 , several loci reported as genome-wide significant in that manuscript are not significant in our study. Examples of these are variants in loci nominated as HESX1, HS3ST1, BZRAP1-AS1, SUZ12P1, and ALPK2. Interestingly, in a follow-up publication to the original GWAS 4 , the same group did not find associations at these signals either, after increasing their case sample size by 25% and their controls by approximately threefold. This was true despite using the same base data published in the initial GWAS. Had we removed these variants from our analysis, as Uffelmann et al. suggest, we would not have been able to determine that they were false positives in the base GWAS and, more generally, would have missed the possibility of auditing the originally published loci....