coranova

This package uses sample correlations to formally compare two or more polygenic scores in one or more populations.
Two implementations are available: parametric and non-parametric. We recommend the parametric implementation for continuous outcomes, and non-parametric for binary outcomes, or other non-normally distributed traits.

Paper forthcoming.

Installation

You can install the development version of coranova from GitHub with:

# install.packages("devtools")
devtools::install_github("gunns2/coranova",  build_vignettes = TRUE)
library(coranova)

Once installed, you can access detailed examples of how to use the package in our two vignettes with

browseVignettes("coranova")

Quick Start

Data Preparation

To compare polygenic scores across multiple population samples, create a list of data frames where each data frame contains the data from a distinct population sample. Each data frame should contain a column with the outcome variable and computed polygenic scores to be compared. The names of the columns should be shared across population sample data frames.

We can use this package to compare the performance of three polygenic scores in two populations. For full example, please see example1.Rmd.

AFR:

  pheno       pgs1        pgs2        pgs3
  -1.4606535  -1.1363873  0.7988800   0.05945586
  -0.1498208  0.9218904   -1.8164597  -1.95811704
  3.5076853   0.2239545   1.5862969   -0.12250429

EUR:

  pheno       pgs1        pgs2        pgs3
  -3.249809   0.66336192  -2.470082   -0.5937651
  -2.754595   -0.72927645 2.431823    -0.1291377
  -1.698634   0.07946734  -1.129588   -0.9772898

To compare all three PGS across the two populations:

perform_coranova_parametric(list(afr, eur), "pheno", c("pgs1", "pgs2", "pgs3"))

To compare pgs1 and pgs2 in African population:
This command will give confidence interval for difference in correlation between outcome and the two scores in the African sample.

perform_coranova_parametric(list(afr), "pheno", c("pgs1", "pgs2"))

Warnings

Both PGS and outcome should be adjusted for principal components prior to analysis.

Prior to running any comparisons, ensure that all correlations between polygenic scores and outcomes are positive. If they are not, the polygenic score was likely miscalculated. This can happen if the polygenic score was computed with the incorrect effect allele, which in that case it is a simple fix to multiple the polygenic score by -1. However, it is important to check that this is the problem and not a larger problem with the score.

In simulations we find at least 1000 bootstrap and 1000 permutations are necessary to control type 1 error with binary outcomes. We have not tested other non-normal outcome distributions.

References

Bilker WB, Brensinger C, Gur RC. A Two Factor ANOVA-like Test for Correlated Correlations: CORANOVA. Multivariate Behav Res. 2004 Oct 1;39(4):565-94. doi: 10.1207/s15327906mbr3904_1. PMID: 26745459.

Olkin I, Finn JD. Testing correlated correlations. Psychological Bulletin. 1990 Sep;108(2):330.

Contact

Sophia Gunn gunns2@bu.edu