Hi,

I am trying to run Limma's LmFit() and eBayes() on a ESET merged using InSilicoMerging. Both functions are giving me errors. The Heart_Failure column contains the values yes, no and NA. 

Here is my code:

eset57338 <- (getDataset("GSE57338", "GPL11532",format = "CURESET", norm = "SCAN", features = "GENE"))

eset42955 <- (getDataset("GSE42955", "GPL6244", format = "CURESET",norm = "SCAN", features = "GENE"))

thelist <- list(eset57338, eset42955)

#Merge datasets

eset_Combat <- merge(thelist, method = "COMBAT")

#Remove NA values in Heart_Failure column

eset_Combat$Heart_Failure[eset_Combat$Heart_Failure == "NA"] <- NA    

keep <- !is.na(eset_Combat$Heart_Failure)

neweset_Combat <- eset_Combat[,keep]

#Create design matrix

new.design <- model.matrix(~ Heart_Failure, pData(neweset_Combat))

#Run lmFit

new.fit <- lmFit(neweset_Combat, new.design) 

e4 <- eBayes(new.fit)​

The lines that are giving me errors:

new.fit <- lmFit(neweset_Combat, new.design) 

Coefficients not estimable: Heart_Failureyes

and

e4 <- eBayes(new.fit)​

Removing intercept from test coefficients

Error in object$cov.coefficients[j, j, drop = FALSE] : 

  subscript out of bounds

In addition: Warning messages:

1: Partial NA coefficients for 17342 probe(s) 

2: In ebayes(fit = fit, proportion = proportion, stdev.coef.lim = stdev.coef.lim,  :

  Estimation of var.prior failed - set to default value

Thanks

As far as I can tell, GSE42955 doesn't have any heart failure phenotypic data, so you are in effect combining data only to exclude one of the sets you are combining. Or am I missing something?

If you want to fit a model on phenotypic data that only exists for one experiment, there is no need to combine. You could use GEOquery and limma directly.