Handling NA values in limma's voom design matrix

Question

[rubic]

Hi, I'm trying to use limma's voom for RNA-seq data, where my design matrix contains NA values. Can voom handle these NAs or is the only solution to toss out the samples with these NAs?

Here's my example data:

set.seed(1)
counts.mat <- matrix(as.integer(runif(1000*10, 10,1000)),nrow = 1000, ncol = 10, dimnames = list(paste0("g",1:1000),paste0("s",1:10)))
design.df <- data.frame(id = paste0("s",1:10),
                        sex = sample(c("female","male"), 10, replace = T),
                        age = sample(c(5, 7, 10), 10, replace = T))
#adding NAs
design.df$sex[3] <- NA
design.df$age[5] <- NA

design.mat <- model.matrix(~ id + age + sex, model.frame(~ ., design.df, na.action = na.pass))
dge <- edgeR::DGEList(counts = counts.mat)
dge <- edgeR::calcNormFactors(dge)
voom.obj <- limma::voom(dge, design.mat, plot=TRUE)

The last last gives this error message:

Error in qr.default(x) : NA/NaN/Inf in foreign function call (arg 1)

Answer 1

[Kevin Blighe]

Cross-posted and answered by the limma developer: https://support.bioconductor.org/p/9144237/