filter out low expressed RNA-seq genes using anova

Question

[jfertaj]

Hi all,

I am analysing RNA-seq data, 37223 genes and 150 samples (RPKM values). I would like to do some co-expression analysis using WGCNA, however I want to filter some genes out with low expressed values first but using some kind of statistical test. I think Partek uses ANOVA and I would like to do the same. I found some code in internet but I don't understand the meaning of dpa

Code:

dpa=as.factor(rep(rep(10,40,by=10),each=3))
fanova <- function(x){anova(aov(x~dpa))$"Pr(>F)"[1]}
result <- apply(dataRNA, 2, fanova)

The example code is for 12 RNA-seq samples (also RPKM values)

Thanks in advance

[komal.rathi]

Can you post the link to the source of this code?

Answer 1

[mark.ziemann]

You don't need to run Anova to do non-differential filtering for RNA-seq data. There are generally 2 approaches:

1. Discard a set proportion of genes (say 30%) or
2. Even better, discard genes with less than a certain number of reads per sample (across the whole experiment). I use 10 reads per sample as a rule of thumb that works.

Also, no need to use FPKM. The standard tools for RNA-seq DGE use count (integer) data; i.e., edgeR & DESeq.

[jfertaj]

Thanks Mark, however I don't have the count data, only RPKM. Also I am not interested in doing DGE, I have data from different tissues but with only one condition (normal tissue). What I would like to do is to create some co-expression networks, one for each tissue, and I think it would be better to remove some low expressed genes. The data is log2(RPKM), some people suggest to remove those genes with log2(RPKM) < 2 but I would like to use some "statistical approach" to do it