Normalization / Transformation of DE data with a de novo assembly

[cement_head]

Hello,

I have a question regarding transformation and/or normalization as it applies to analysis of differential expression (DE) for a non-model (novel) de novo assembled transcriptome.

(1) I have an assembled transcriptome that I'm happy with - let's not get into that

(2) For DE analysis, does one have to do both a transformation and normalization, or can one simply do one or the other - depending on the downstream questions / analysis? Or are these two totally different things that can be mutally exclusive?

(3) For DE analysis, it seems (and here's where my understanding is sketchy) that some of the normalization methods are better suited to organisms whose genomes are gene annotated. Because I'm using the circular approach of making a transcriptome from the reads and then using that (transcriptome) as the reference, some of the approaches to normalization would likely (maybe) not apply. I know that RSEM used in conjunction with Trinity (Broad Institute) uses TMM. Would RPKM also (not in addition to) still be a relavent normalization approach for DE analysis of transcriptome data? Or would RPKM inadvertently up the small transcripts?

Thanks,
Andor

[westerman]

Since denovo transcript followed by mapping/analysis with the same reads contributing to the denovo assembly is something I do all of the time I am curious what other people have to say on this issue.

My feeling is that it does not matter -- either method should be fine. But then with denovo work I am often just happy to find something out at the gross (or top) level that we can chew on and do not get concerned about finer details that might interest a well-characterized genome project such as a Human study. In other words you may be focusing in on a detail of transcriptome analysis that won't really matter that much given the data you have.

[cement_head]

Since denovo transcript followed by mapping/analysis with the same reads contributing to the denovo assembly is something I do all of the time I am curious what other people have to say on this issue.

My feeling is that it does not matter -- either method should be fine. But then with denovo work I am often just happy to find something out at the gross (or top) level that we can chew on and do not get concerned about finer details that might interest a well-characterized genome project such as a Human study. In other words you may be focusing in on a detail of transcriptome analysis that won't really matter that much given the data you have.

yeah, good point - thanks.

[cement_head]

Hello,

I have a question regarding transformation and/or normalization as it applies to analysis of differential expression (DE) for a non-model (novel) de novo assembled transcriptome.

(1) I have an assembled transcriptome that I'm happy with - let's not get into that

(2) For DE analysis, does one have to do both a transformation and normalization, or can one simply do one or the other - depending on the downstream questions / analysis? Or are these two totally different things that can be mutally exclusive?

(3) For DE analysis, it seems (and here's where my understanding is sketchy) that some of the normalization methods are better suited to organisms whose genomes are gene annotated. Because I'm using the circular approach of making a transcriptome from the reads and then using that (transcriptome) as the reference, some of the approaches to normalization would likely (maybe) not apply. I know that RSEM used in conjunction with Trinity (Broad Institute) uses TMM. Would RPKM also (not in addition to) still be a relavent normalization approach for DE analysis of transcriptome data? Or would RPKM inadvertently up the small transcripts?

Thanks,
Andor

Found this paper here: http://www.ncbi.nlm.nih.gov/pubmed/22988256

in a nutshell, use RAW counts as input data for DE analysis. Don't use RAW counts or RPKM (FPKM) for between sample comparisons...ever.