How To Reduce The Number Of Genes For A Microrray Experiment Done On Cell Lines (Dimensionality Reduction)?

Question

[Hmm]

I have a dataset with 18 cell lines (HG_U133_plus2) normalized using GCRMA. I have collapsed the dataset to genes using the median of the probes. Before proceeding further with any kind of differential expression, how do you guys deal with more than 20 thousand genes.

1. Do you remove genes that have low variance across all cellines? If so then do the genes that are highly expressed across all cell lines will be removed...am i on the right track.
2. How do you select the threshold for a variance across cell lines per gene?

Thanks

Answer 1

[ewre]

Haven't analyzed cell line micro-arrays, but have analyzed micro-arrays examining cells extracted from blood. usually after background correction and normalization, i directly carry out DEGs analysis without gene expression variance filtration, 20k genes is not very much for DEG analysis tools. But if you want to do co-expression analysis, it is better to filter out genes that are of small expression variance for 2 reasons:

1. genes with small expression variance usually get extremely high co-expression values which can be deceiving
2. decrease the total # of input genes will accelerate the calculation of correlation matrix and subsequent analysis steps(cluster and network analysis)

besides: filtering out unrelated genes is also a good idea when you do pca analysis.