Feature selection for constructing gene regulatory networks from microarray data

Question

[bioinf]

Hi,

I'm trying to construct a gene regulatory network from gene expression data. I've got around 200 samples (breast tumour) with expression values for 25k genes. Every network inference method I've tried so far (simple correlation, genie3, aracne) seems to require a smaller number of genes to work with.

I've tried selecting top N (10 to 200) differentially expressed genes and working with them, but edge confidence values were quite low. In addition, when looking at the correlation matrix, there are almost no negative values (i.e. no repression) and very few positive correlation values above 0.5. This makes me believe that differential expression might not be the best strategy for selecting important genes.

I would really appreciate any suggestions for a better feature selection approach.

Thanks.

[Xiaokang]

Hi,

I'm recently working on selecting the important genes using both statistical methods and feature selection methods, and making a comparison between them.

But I can't understand your question. What do you mean by "edge confidence"? And why does "almost no negative values (i.e. no repression) and very few positive correlation values above 0.5" makes you believe that "differential expression might not be the best strategy for selecting important genes"? I can't find the logic in it. Could you please explain that a bit?

Thx.