MCANOVA Package

The MC-ANOVA R package provides:

• MC_ANOVA: A function to estimate the Relative Accuracy (RA) of cross-ancestry prediction for short chromosome segments.

• Maps of the Relative Accuracy of European-derived local genomic scores for African, Caribbean, East Asian, and South Asian ancestry groups.

• A Shiny App providing a graphical interface to the Relative Accuracy maps. (Note, a slower but URL-acccessible version of the app can be found here)

• Tools that, together with MC_ANOVA() can be used develop Relative Accuracy maps.

• Link to manuscript

Installation

To install the development version from GitHub, first install remotes:

 install.packages("remotes")
 library(remotes)

Then install the package:

 # detach("package:MCANOVA", unload = TRUE)
 install_github("lupiA/MCANOVA")
 library(MCANOVA)

This should take about one minute for a typical computer when including installation of all non-base dependencies.

This package should be compatible with Windows, Mac, and Linux operating systems and has been tested on Windows 7 & 10, macOS Monterey, and Linux CentOS 7.

Examples

• Loading Relative Accuracy maps in an R session.
• Shiny App: Launches a Shiny App for the RA Maps we developed using UK Biobank data.
• Segments: Finds disjoint chromosome segments.
• MCANOVA: Estimate Within- and Cross-ancestry R-squared.

Loading the Relative Accuracy maps into an R session

 library(MCANOVA)

# Relative Accuracy Map (UK Biobank arrays)
 data(MAP_UKB)

Back

Launching the Shiny App

 # library(ggplot2)
 # library(shiny)

 library(MCANOVA)
 PGS_portability_app()

Back

Creating chromosome segments of a minimum base pair length and size (# of SNPs).

 library(MCANOVA)

# Genotype map
 data(geno_map_example)

# Initialize MAP and define segments
 minSNPs <- 10
 minBP <- 10e3
 MAP_example <- geno_map_example
 MAP_example$segments <- getSegments(MAP_example$base_pair_position, chr = MAP_example$chromosome, minBPSize = minBP, minSize = minSNPs, verbose = TRUE)

Back

Running MC-ANOVA

This example requires the R package BGData which is installed along with the MCANOVA package:

# Load necessary packages
# install.packages("BGData")
# library(BGData)
 library(MCANOVA)
 library(BGData)

# Set seed
 set.seed(12345)

# Generate genotypes (100 subjects and 500 SNPs)
 n <- 100
 p <- 500
 X <- matrix(sample(0:2, n * p, replace = TRUE), ncol = p)
 data(geno_map_example)
 colnames(X) <- geno_map_example$SNPs
 minSNPs <- 10
 minBP <- 10e3
 MAP_example <- geno_map_example
 MAP_example$segments <- getSegments(MAP_example$base_pair_position, chr = MAP_example$chromosome, minBPSize = minBP, minSize = minSNPs, verbose = TRUE)

# Assign ancestry IDs (80% to ancestry 1, 20% to ancestry 2)
 n_1 <- round(0.8 * n)
 n_2 <- round(0.2 * n)
 ancestry <- rep(c("Group_1", "Group_2"), times = c(n_1, n_2))
 rownames(X) <- ancestry

# Initialize portability estimates
 MAP_example$correlation_within <- NA
 MAP_example$correlation_across <- NA

# Set parameters for MC-ANOVA

# a small constant added to the diagonals of X'X to avoid numerical errors when some SNPs are in perfect LD
 lambda <- 1e-8
# number of Monte Carlo simulations
 nRep <- 300
# number of causal variants
 nQTL <- 3

# Loop over segments and run MC-ANOVA
 for (i in min(MAP_example$segments):max(MAP_example$segments)) {
   core <- which(MAP_example$segments == i)
   flank_size <- 10
   chunk_start <- max(min(core) - flank_size, 1)
   chunk_end <- min(max(core) + flank_size, nrow(MAP_example))
   chunk <- chunk_start:chunk_end
   isCore <- chunk %in% core
  
   X_1 <- X[rownames(X) == "Group_1", chunk]
   X_2 <- X[rownames(X) == "Group_2", chunk]
  
   # Run MC-ANOVA
   out <- MC_ANOVA(X = X_1, X2 = X_2, core = which(isCore), lambda = lambda, nQTL = nQTL, nRep = nRep)
  
   # Extract portability estimates
   MAP_example$correlation_within[chunk[isCore]] <- out[1, 1]
   MAP_example$correlation_across[chunk[isCore]] <- out[2, 1]
 }

 RA <- MAP_example$correlation_across^2/MAP_example$correlation_within^2

Back

Expected outputs

• MAP_example: a 500 x 10 data frame with columns 1-3 containing variant information (chromosome, RS ID, and base pair position), column 4 containing the numeric segment the SNP belongs to estimated from getSegments(), column 5 containing the within-ancestry group MC_ANOVA() correlation estimates, and columns 6 containing the across-ancestry group MC_ANOVA() correlation estimates.
• RA: a length 500 vector of Relative Accuracy estimates.
• Interactive shiny app interface.

All demos should take only a few seconds to run.

System Requirements

Depends: R (>= 3.5.0) Imports: shiny, ggplot2, BGData

References

• Wang et al.(Nat. Comm., 2020) Theoretical and empirical quantification of the accuracy of polygenic scores in ancestry divergent populations.
• Lupi, A.S., Vazquez, A.I. & de los Campos, G. Mapping the relative accuracy of cross-ancestry prediction. Nat Commun 15, 10480 (2024). https://doi.org/10.1038/s41467-024-54727-8