For estimating batch effects using SVA, perform the following steps:

1. Load the libraries and data:

library(sva)
arab <- readRDS(file.path(getwd(), "datasets", "ch1", "arabidopsis.RDS"))

2. Filter out rows with too few counts in some experiments:

keep <- apply(arab, 1, function(x) { length(x[x>3])>=2 } )
arab_filtered <- arab[keep,]

3. Create the initial design:

groups <- as.factor(rep(c("mock", "hrcc"), each=3))

4. Set up the test and null models and run SVA:

test_model <- model.matrix(~groups)
null_model <- test_model[,1]
svar <- svaseq(arab_filtered, test_model, null_model, n.sv=1)

5. Extract the surrogate variables to a new design for downstream use:

design <- cbind(test_model, svar$sv)