[BioC] Re: problem with LPE

[Nitin Jain]

Dear Philippe,

As the warning messages indicate, NANs are being produced
in taking square root, which means some values in var.x are being
calculated as negative from baseOlig.error function. I apply preprocess
function to the data set, before giving it as input for baseOlig.error
function, which takes care of missing values, thresholds the data at
1, and does log2 transformation.

Could you please send me the data4, so that I can check what
part is going wrong?

Thanks.

Nitin
PhD candidate, University of Virginia
------------------------------------


> Hi Nitin and all,
> I m trying to use LPE to compare 2 groups. In first group I have 4 chips and in the second one 8 chips (same kind of CEL files, ie, HU133A from Affymetrix).
> here is what I doing:
>
> library(affy)
> library(LPE)
> library(gcrma)
>
> data1<-ReadAffy()
> data2 <-gcrma(data1)
> data3<-exprs(data2)
> data4<-data.frame(Name=geneNames(data2), data3)
> var.g1 <- baseOlig.error(data4[,2:5],q=0.01)   # group1 colum 2 a 5
> var.g2 <- baseOlig.error(data4[,6:13],q=0.01)   # group2 colum 6 a 13
>
> lpe.val1<-data.frame(lpe(data4[,6:13], data4[,2:5], var.g2, var.g1, probe.set.name=data4$Name))
>
> And here is what I obtain first:
> Warning messages:
> 1: NaNs produced in: sqrt(1.57 * ((var.x/n1) + (var.y/n2)))
> 2: NaNs produced in: sqrt(var.x)
> 3: NaNs produced in: sqrt(var.y)
>
> then the following:
>
> > lpe.val <- round(lpe.val1, digits=2)
> > fdr.BH <- fdr.adjust(lpe.val, adjp="BH")
> > fdr.2 <- fdr.adjust(lpe.val, adjp="resamp", iterations=2)
> iteration number 1 is in progress
> iteration number 1 finished
> iteration number 2 is in progress
> iteration number 2 finished
> Computing FDR...
> Error in "[<-"(`*tmp*`, j, value = z.real[num.genes.signif]) :
>         nothing to replace with
> In addition: Warning messages:
> 1: NaNs produced in: sqrt(1.57 * ((var.x/n1) + (var.y/n2)))
> 2: NaNs produced in: sqrt(var.x)
> 3: NaNs produced in: sqrt(var.y)
>
> When I look at the file I dont get any adjusted p values
> any suggestion ?
> thanks
> Philippe
>
> ps I ve been also trying the old version of LPE without permutation
> lpe.val1<-data.frame(lpe(data4[,6:13], data4[,2:5], var.g2, var.g1, probe.set.name=data4$Name, adjp=c("BH","BY")))
>
> and obtained the same warning meesages:
> 1: NaNs produced in: sqrt(1.57 * ((var.x/n1) + (var.y/n2)))
> 2: NaNs produced in: sqrt(var.x)
> 3: NaNs produced in: sqrt(var.y)
> and no adjusp P values
>
>