Microarray Data-> Filtering Genes

Question

[pixie@bioinfo]

I am tying to filter some genes after normalizing them. I came across two methods in dchip. One using a CV cutoff (standard deviation/mean) and the other Percentile Filtering technique. I could not find my explanation regarding this, especially Percentile Filtering. When do we apply which method and what is the outcome ?...my datasets corresponds to plant in normal and drought condition and I want to cluster the genes using WGCNA. Any suggestion would be greatly appreciated.

Answer 1

[Devon Ryan]

Normally, one filters after normalization and before calculating differential expression to perform fewer tests. While this also speeds things up, the main benefit is due to not then having to correct for performing a bunch of tests that could never possibly have given significant results (this is why you need to filter carefully).

For WGCNA, I never tried filtering my data prior to use (granted, I have RNAseq rather than microarray data). Aside from faster computation (since you'll have smaller matrices), I'm not sure there's a big benefit to filtering with WGCNA.

[Rm]

@Devon Ryan: for RNASEQ input to WGCNA, do you use counts or fpkm or fold change as input? ....thanks

[Devon Ryan]

One of the authors has recommended variance stabilized data, so I'd go with that.

[Rm]

thanks @Devon Ryan: