Calculate RPKM of replicates

[nelsonwd]

I can't find anything through searching so this must be a pretty noob question but any comments are appreciated.
I want to display abundance in expression experiments in a common format regardless of the tool used for the analysis (i.e. EdgeR, DESeq).
We have chosen RPKM to represent abundance using the formula:
(number of reads/kb of exon)/mill mapped reads

When there are multiple replicates how are the runs combined?:
1) the total count from all of the replicates (and the sum of the mapped reads from all of the replicates).
2) the average count of the replicate counts (The sum of the average counts used for the number of mapped reads)
3) the base mean of the replicate counts (the sum of the base mean used as the of mapped reads)

Thanks,
Bill

[swbarnes2]

I would think you would want the average and the standard deviation.

You do replicates so that you know how much variation is natural, so you can detect exceptional levels of variation.

[nelsonwd]

Thanks for the reply. This doesn't seem as common as I thought it was. Talking to other people they calculate a separate RPKM for each replicate and don't combine them. I want an RPKM version of DESeqs base mean.
Wouldn't the standard deviation be redundant to the p-value, and fdr?

I'm starting to think that calculating RPKM from averaged counts, averaged RPKMs, or from the base mean are all pretty much the same.

[chadn737]

What kind of replicates are they? What specifically is the purpose?

[nelsonwd]

There are four conditions. Two technical replicates per condition. The purpose is differential expression.