Get Normalized Read Count Sample Matrix Tophat/Cufflinks

Question

[Irsan]

I have sequenced RNA from 12 human samples (6 tumor (of which 3 tumor group A and 3 tumor group B), 6 matched non-tumor) samples. Using tophat, I have aligned the reads to hg19 and with cufflinks I have made transcript models for each sample. I would like to extract the FPKM values for each sample in matrix format so that I can do hierarchical clustering and principal compononent analysis on all 12 samples.

The problem is that for each sample, different transcripts are assembled by cufflinks so I cannot just paste the cufflinks files together to get the matrix. Something that came into my mind to do this was using a reference transcript file and use bedtools/bedops to look for intersecting transcripts in all 12 samples. However, I hope I am overlooking some functionality in cufflinks/cuffcompare/cuffdiff to get this done more easily

Answer 1

[biopaw]

CummeRbund R package may be what you need. A colleague of mine uses this (I don't), and it continues the workflow, creates a database with the outputs from Tophat/Cuffdiff and implements several plotting functions. When using the tophat suite the workflow is more tightly controlled (hence CummeRbund), which may be great for the casual user

But I can recommend that you may be better off using a STAR (if you have a puter with at least 36G RAM, you need 16G for Human index), HTseq for counting, and EdgeR (there are other good R seq packages as well). Your RNA-seq alignment would be completed in a few minutes, instead of a few hours, and you trade in the convenience of the more rigid workflow for a more flexible one in R (more work), but you can take advantage of the other Packages in Bioconductor.

With Edge, you could model the effect directly A vs B by creating a contrast A[TNT] - B[T-NT], where as in Cuffdiff, it seems you can only model direct, T vs NT for example. Then you can also perform any plot you like usein the Bioconductor tools in R, so you can then do the hierarchical clustering, PCA, MDS etc; ggplot is a nice plotting tool in R.

P

[Irsan]

Thanks, I will try STAR + HTSEQ + EdgeR next to tophat + cufflinks + cuffdiff

Answer 2

[Ryan Thompson]

You probably want to use cuffmerge to combine all the individual sample assemblies, and then re-run each sample using the merged assembly as a reference.

[Irsan]

That sounds like something I am looking for indeed :-)

[Irsan]

Thanks Ryan, I ran cufflinks with --GTF (not --GTF-guide) with the transcripts.gtf from cuffmerge and it worked like a charm :-)