Hi

I am about to analyse a load of omic data from multiple platforms and it seems sensible for feature selection to use the same statistical test for all data types. Our response variable is continuous, but it is also discrete in another analysis. I was wondering what the best way of doing this was? I usually use limma for array and RNA-seq, but having to analyse these other data probably requires a different approach.

Our analysis includes flow cytometry, RNA-seq, microarray, proteomics, and metabolomics.

Thanks,

Chris

Your question is pretty vague, so I'll give an equally vague answer. In general, if I have data sets of many different types, I start off by analyzing each one on its own merits. Only after I obtain results from each analysis do I think about how to integrate them, e.g., by overlapping differentially bound sites from ChIP-seq with promoters of DE genes or by intersecting DE lists from multiple transcriptomic experiments. In all cases, I pick what I think is the best analysis and/or statistical method for each data set; this ensures that I get reliable results for integration. I don't concern myself with the fact that the analyses might be different between the different data types. Why would that matter, as long as power is good and FDR control is maintained in each analysis?

You mention four platforms. You're already using limma for the microarray and for the RNA-seq, and limma is also a common choice for proteomics. So what platform is causing you a problem?

I don't know what you mean by a "flow cytometry" platform, because flow cytometry is a method for sorting cells rather than a genomic profiling technology.