Entering edit mode
everardoremi
•
0
@everardoremi-23233
Last seen 4.7 years ago
Hello everyone, I am trying to run a differential expression analysis with duplicates, but do I get this warning message any suggestions?
# Create identifiers for case-control status
Group <- factor(targets$Cy5, levels = c("normal", "cancer"))
# Generate design matrix including case-control status and analytical batch
design <- model.matrix(~ Group)
colnames(design) <- c("NormalvsRef","CancervsNormal")
# Estimate array quality weights:
aw <- arrayWeights(MA_norm,design)
# The array weights multiply the spot weights already in the data object:
w <- matvec(MA_norm$weights,aw)
# Account for duplicate spots
dupcor <- duplicateCorrelation(MA_norm,design,ndups=2)
# Fit linear model to the normalized data using the design matrix and including batch as covariate
fit <- lmFit(MA_norm, design=design,ndups=2,correlation=dupcor$consensus, weights = w)
Warning message:
Partial NA coefficients for 10 probe(s)
Yes, it's caused by NAs. The message means that for 10 of the probes, all the expression values for one of the groups (either normal or cancer) are NA (or had zero weight, which is the same thing). If either of the groups are entirely NA then obviously the log-fold-change between the groups must be NA as well, which leads to the message.
0 variance doesn't actually cause any problems. limma can handle that one without introducing NA coefficients.
At this stage, I always add the extra note -- does OP need to have NA expression values? In most cases, NA expression values can be avoided by appropriate normalization and background correction.