Recent demography and natural selection hamper power of rare variant association tests. L. H. Uricchio1, J. S. Witte2,3, R. D. Hernandez3,4,5 1) Joint Bioengineering Graduate Group, UCSF and UC Berkeley, San Francisco, CA; 2) Department of Epidemiology and Biostatistics, UCSF, San Francisco, CA; 3) Institute for Human Genetics, UCSF, San Francisco, CA; 4) Institute for Quantitative Biosciences, UCSF, San Francisco, CA; 5) Bioengineering and Therapeutic Sciences, UCSF, San Francisco, CA.
Current findings suggest that common genetic variants may explain only a fraction of the heritability of complex human diseases, and recent research focus has turned toward rare variants. However, rare variants can only explain a large proportion of the additive genetic variance of complex traits when they have dramatically larger effect sizes than common variants. The simplest explanation for an inverse relationship between allele frequency and effect size is that natural selection prevents trait altering alleles from increasing in frequency. Unfortunately, most statistical tests for association between rare alleles and complex traits do not explicitly model natural selection. Moreover, recent demographic effects--such as the explosive growth experienced across many human populations--can also influence the genetic architecture of complex traits. To evaluate the impact of natural selection and demography on rare variant association tests, we jointly simulate DNA sequences and complex phenotypes under recently inferred models of human selection and multi-population demographic effects. We find that the statistical power of state-of-the-art rare variant tests (e.g., SKAT-O) is a decreasing function of the correlation between selection strength and effect size. This counterintuitive result means that rare variant-based statistical tests perform worst when rare variants explain a large proportion of the additive variance, even for large sample sizes in the thousands. We investigate the sensitivity of this conclusion to the length of the causal locus, the total number of causal loci, the shape of the distribution of selection coefficients and effect sizes, and sample size. Finally, we show that population genetic predictions of the relationship between allele frequency and the fraction of phenotypic variance explained could be used to characterize the distribution of effect sizes for complex human phenotypes.
You may contact the first author (during and after the meeting) at