symDMatrix is an R package that provides symmetric matrices partitioned into file-backed blocks.
A symmetric matrix G
is partitioned into blocks as follows:
+ --- + --- + --- +
| G11 | G12 | G13 |
+ --- + --- + --- +
| G21 | G22 | G23 |
+ --- + --- + --- +
| G31 | G32 | G33 |
+ --- + --- + --- +
Because the matrix is assumed to be symmetric (i.e., Gij
equals Gji
), only the diagonal and upper-triangular blocks are stored and the other blocks are virtual transposes of the corresponding diagonal blocks. Each block is a file-backed matrix of type ff_matrix
of the ff package.
The package defines the class and multiple methods that allow treating this file-backed matrix as a standard RAM matrix.
Before we start, let’s create a symmetric matrix in RAM.
library(BGLR)
# Load genotypes from a mice data set
data(mice)
X <- mice.X
rownames(X) <- paste0("ID_", 1:nrow(X))
# Compute a symmetric genetic relationship matrix (G matrix) in RAM
G1 <- tcrossprod(scale(X))
G1 <- G1 / mean(diag(G1))
In practice, if we can hold a matrix in RAM, there is not much of a point to convert it to a symDMatrix
object; however, this will help us to get started.
Now that we have a symDMatrix
object, let’s illustrate some operators.
# Basic operators applied to a matrix in RAM and to a symDMatrix
# Dimension operators
all.equal(dim(G1), dim(G2))
nrow(G1) == nrow(G2)
ncol(G1) == ncol(G2)
all.equal(diag(G1), diag(G2))
# Names operators
all.equal(dimnames(G1), dimnames(G2))
all(rownames(G1) == rownames(G2))
all(colnames(G1) == rownames(G2))
# Block operators
nBlocks(G2)
blockSize(G2)
# Indexing (can use booleans, integers or labels)
G2[1:2, 1:2]
G2[c("ID_1", "ID_2"), c("ID_1", "ID_2")]
tmp <- c(TRUE, TRUE, rep(FALSE, nrow(G2) - 2))
G2[tmp, tmp]
head(G2[tmp, ])
# Exhaustive check of indexing
for (i in 1:100) {
n1 <- sample(1:50, size = 1)
n2 <- sample(1:50, size = 1)
i1 <- sample(1:nrow(X), size = n1)
i2 <- sample(1:nrow(X), size = n2)
TMP1 <- G1[i1, i2, drop = FALSE]
TMP2 <- G2[i1, i2, drop = FALSE]
stopifnot(all.equal(TMP1, TMP2))
}
The function getG_symDMatrix
of the BGData package computes G=XX’ (with options for centering and scaling) without ever loading G in RAM. It creates the symDMatrix
object directly, block by block. In this example, X
is a matrix in RAM. For large genotype data sets, X
could be a file-backed matrix, e.g., a BEDMatrix
or ff
object.
library(BGData)
G3 <- getG_symDMatrix(X, blockSize = 400, vmode = "double", folderOut = "mice2")
class(G3)
all.equal(diag(G1), diag(G3))
for(i in 1:10){
n1 <- sample(1:25, size = 1)
i1 <- sample(1:25, size = n1)
for(j in 1:10){
n2 <- sample(1:nrow(G1), size = 1)
i2 <- sample(1:nrow(G1), size = n2)
tmp1 <- G1[i1, i2]
tmp2 <- G3[i1, i2]
stopifnot(all.equal(tmp1, tmp2))
}
}
ff
files containing the blocksThe function symDMatrix
allows creating a symDMatrix
object from a list of .RData
files containing ff_matrix
objects. The list is assumed to provide, in order, files for G11, G12, ..., G1q, G22, G23, ..., G2q, ..., Gqq
. This approach is useful for very large G matrices. If n
is large it may make sense to compute the blocks of the symDMatrix
object in parallel jobs (e.g., in an HPC). The function getG
of the BGData package is similar to getG_symDMatrix
but accepts arguments i1
and i2
which define a block of G (i.e., rows of X
).
library(BGLR)
library(BGData)
library(ff)
# Load genotypes from a wheat data set
data(wheat)
X <- wheat.X
rownames(X) <- paste0("ID_", 1:nrow(X))
# Compute G matrix in RAM
centers <- colMeans(X)
scales <- apply(X = X, MARGIN = 2, FUN = sd)
G1 <- tcrossprod(scale(X, center = centers, scale = scales))
G1 <- G1 / mean(diag(G1))
# Compute G matrix block by block (each block computation can be distributed)
nBlocks <- 3
blockSize <- ceiling(nrow(X) / nBlocks)
i <- 1:nrow(X)
blockIndices <- split(i, ceiling(i / blockSize))
for (r in 1:nBlocks) {
for (s in r:nBlocks) {
blockName <- paste0("wheat_", r, "_", s - r + 1)
block <- getG(X, center = centers, scale = scales, scaleG = TRUE,
i = blockIndices[[r]], i2 = blockIndices[[s]])
block <- ff::as.ff(block, filename = paste0(blockName, ".bin"), vmode = "double")
save(block, file = paste0(blockName, ".RData"))
}
}
G2 <- as.symDMatrix(list.files(pattern = "^wheat.*RData$"))
attr(G2, "centers") <- centers
attr(G2, "scales") <- scales
all.equal(diag(G1), diag(G2)) # there will be a slight numerical penalty
Install the stable version from CRAN:
Alternatively, install the development version from GitHub:
Further documentation can be found on RDocumentation.