RNASeq- Which method appropriate for Normalization?

[daanum]

Hi,
This question is to the people who have worked with RRBS methylation data and RNASeq data. I have RRBS methylation data for cell lines and the RNASeq data for the same cell lines. For RNASeq data I have used cuffdiff program for differential expression analysis but I am not quiet happy with it as it does not give per sample fpkm or read count information so that I can compare the expression in groups with the expression of each sample in those groups to validate the results of differential expression analysis. I would like to use other method of differential expression but I have a question that since I will be integrating the data with RRBS, should I consider only the method which generates the read counts in fpkm units(like cufflinks-cuffdiff)? I am thinking this because fpkm normalizes the reads by transcript length which other methods do not.

My second question is that is there a standard method to integrate RRBS methylation data and RNASeq paired end data?

Any suggestions will be greatly appreciated. Thanks.

[bastianwur]

Why would the length matter? You should be comparing the same gene between conditions (then the length doesn't matter). Comparing different genes in the same condition (where a length correction would be necessary) doesn't make sense anyways (different kinetics of the proteins, different regulation/repression, different protein half life, etc.).

For a read on normalization methods, this paper http://www.ncbi.nlm.nih.gov/pubmed/22988256 is the one to go to.
Use DESeq2 or EdgeR (as short conclusion).

[Macspider]

Use DESeq2 or EdgeR (as short conclusion).

Totally agree. Using FPKM or RPKM in this case can be done because cufflinks outputs it automatically in the isoforms.fpkm_tracking file so you don't bother using other units.
What you need is to do a between lane normalization. This because different experiments can have different sequencing depths and statistical properties, due to the fact that you don't get the results in the same run. Thus, I would recommend a full quantile normalization or a global scaling. This will ensure you that the two lanes in the end have the same max value, mean and variance. Thus, they will be comparable.

EDIT:
I used EdgeR with cufflinks output and full quantile normalization, you can do that with many Bioconductor packages in R. I did it with EDASeq.