Cufflinks bias correction and DESEq2

Question

[as9309]

I ran all my RNA Seq data through galaxy to get FPKM files from original fastq data, but I did bias correction in Cufflinks. However, the DESeq2 manual stresses that you should use raw integer counts, not normalised counts. Does cufflinks normalise the data to some extent during bias correction? Should I have not performed bias correction before doing the DESEq2 analysis?

Thanks.

[Martombo]

as you wrote, DESeq2 works with raw integer counts. this is something different from FPKM values (corrected or not). try using featureCounts or HTSeq-count for it.

[as9309]

Hi, thanks - do you know if there is software on R to do this? I believe HTSeq-count requires python and featureCounts (rsubread) is only compatible with Mac?

[Benn]

Do you mean Windows R? I am afraid not, Rsubread only works in R on mac or linux.

You might wanna install a VM for linux?

[Martombo]

you might try easyRNAseq

[Benn]

This question about how to use cufflink data in DEseq2, edgeR, or limma has been asked so many times before. Do posters even look for answers before asking here?

Answer 1

[andrew.j.skelton73]

Use the following route to do a DESeq2 Analysis: Align fastq files to your reference genome using your favourite splice aware aligner (STAR is probably best), Count using HTSeq_Count, or RSubread, then read that data into DESeq2.

Do not use any output from Cufflinks or the Tuxedo pipeline as a whole to feed into DESeq2, that breaks major assumptions of the data. Tuxedo can produce Pseudo Counts, and that's probably the closest thing you'll get, but even then, don't use them to feed into DESeq2.

Answer 2

[Carlo Yague]

Yes, Cufflink does bias correction on FPKM... but FPKM are not raw integer counts in the first place !

To get the raw counts, you'll need to use either FeatureCounts or HTseq-counts, not Cufflink.