Abstract
Estimating the phenotypic correlations between complex traits and diseases based on their genome-wide association summary statistics has been a useful technique in genetic epidemiology and statistical genetics inference. Two state-of-the-art strategies, Z-score correlation across null-effect single nucleotide polymorphisms (SNPs) and LD score regression intercept, were widely applied to estimate phenotypic correlations. Here, we propose an improved Z-score correlation strategy based on SNPs with low minor allele frequencies (MAFs), and show how this simple strategy can correct the bias generated by the current methods. The low MAF estimator improves phenotypic correlation estimation, thus it is beneficial for methods and applications using phenotypic correlations inferred from summary association statistics.
Highlights
Phenotypic correlation is an essential parameter that helps understand observational correlations between complex traits and the etiological perspectives underlying complex diseases
We show that selecting single nucleotide polymorphisms (SNPs) with low minor allele frequencies (MAFs) can lead to simple and consistent estimation of phenotypic correlations based on multi-SNP Zscore correlations
When there is a non-zero genetic correlation spread across the genome, only those methods that use the SNPs capturing little genetic variance would yield a consistent estimate for the phenotypic correlation, e.g., the random estimator where the SNPs capture absolutely zero genetic variance and the low MAF estimator with low enough MAF cutoffs
Summary
Phenotypic correlation is an essential parameter that helps understand observational correlations between complex traits and the etiological perspectives underlying complex diseases. Estimation of the phenotypic correlation between a pair of phenotypes, by definition, is straightforward in a sample where both phenotypes are measured. Depending on the distribution of each phenotype, the estimated phenotypic correlation serves as a sufficient statistic for many linear statistical models, such as ordinary linear and logistic regressions, allowing us to assess parameters such as odds ratios of risk factors on disease outcomes. Since a large number of genome-wide association studies (GWAS) were conducted, many GWASed phenotypes had measurements in an overlapping set of individuals, where many were from more than one participating cohort in GWAS meta-analysis. Inference of the phenotypic correlations across these phenotypes would be complicated if estimating using the conventional way, which requires individual-level phenotypic data and subsequent meta-analysis. The phenotypic correlations can be estimated based on established GWAS summary statistics, especially when the proportion of sample overlap between two GWASed phenotypes is large. Two state-of-the-art strategies were proposed: Improved Estimation of Phenotypic Correlations
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have