Independent point independence in RNA-seq?

[BGould]

Hi All,
I am in the midst of trying out different methods for measuring the statistical significance of gene expression levels in and RNA seq experiment. It seems like all the methods I come across (parametric and nonparamentric) require that expression measurements (or their errors) be independent of each other.

Sorry if this is a dumb question, but since genes are part of networks, are there not usually high levels of non-independence among expression value measurements in any transcriptome-wide analysis? Does this negate the usefulness of published statistical methods for detecting differential expression?

[Wolfgang Huber]

Dear BGould

It is not true that all or even most methods make sweeping assumptions of independence. In this context, it is helpful to distinguish dependence (i) between the measurements of several genes within one sample, and (ii) between measurements for the same gene, across several samples.

The differential expression methods I am aware of are unaffected by the common types of (i). (ii) should not happen for well-designed experiments and good quality data. If it is a concern, you will need to identify the factor that causes the correlation and model it. Or, refer to the work by Jeff Leek (e.g. http://www.bioconductor.org/packages.../html/sva.html) and similar approaches of that type.

Best wishes
Wolfgang

[BGould]

Thanks Wolfgang!