Hi all,
I have a data.frame from a rna-seq experiment, and I would like to remove some outliers. The data is huge with 350 samples and 32291 genes. The data are log2 RPKM values (I did the log2 because I am planning to do WGCNA analysis and the authors recommend to make a log2 transformation of the data).
I am using the PcaHubert function from rrcov package to find outliers, here is the code I am using:
df <- read.table("/path/to/file/rpkm.txt")
dim(df) #32291 352
df <- df[,-c(1,2)] # first 2 columns have accessory data
library(rrcov)
pcaHub <- PcaHubert(t(df))
outliers <- which(pcaHub@flag=='FALSE')
The outliers would be those samples with the flag FALSE after doing the RobustPCA, do you think it is appropriate to remove outliers using this method?
Any comments would be greatly appreciated
Thanks
If you are going to use WGCNA package for network analysis, than you would be having the option to remove the outliers(samples). Follow the WGCNA Tutorials.
Yes, I think PCA is also a good choice to remove outliers.
you can also hierarchically cluster the samples on spearman's correlation of gene expression. then it would be easy to detect and remove outliers from dendrogram.
Hello There is this parameter "crit.pca.distances" in function PcaHubert what should be the value for this other than default value. And what is this parameter?
Log in to answer this question.
