Suggestions for RNA seq pipeline

Question

[sangita_b]

Hi

I am currently analysing RNA-SEQ data (differential expression). I have 20 samples in total: 5 conditions in 2 donors and each donors- each donor has 2 repeats (cell type- human bronchial epithelial cells).

I have 24fastq files per sample- these have been QC'd in fastqc and run through tophat to produce 1 BAM file per sample.

The BAM files were then run through cufflinks for transcript assembly (gtf files).

I have merged all gtf files in cuffmerge and am not sure whether to use cuffdiff or cuffnorm.

Could I ask if the pipeline I have used so far is correct and also ask for advice how to calculate differentail expression.

Thanks
Sangita

[GouthamAtla]

Do you have annotations available for organism of your interest ?

[sangita_b]

Hi, yes i do.

[GouthamAtla]

Solution given by Tom below would work best.

Answer 1

[Tom]

I do have bad experience using cuffdiff. For human samples and standard differential expression, I personally would recommend you

• Download reference genomes, gtfs from iGenome project
• Align with STAR
• Summarization with htseq-count
• Differentially gene expression with DESeq2 or edgeR

Answer 2

[Antonio R. Franco]

You have many different choices

There is an international consortium that has published a paper comparing several RNA-Seq packages. It run into my attention that they recommend one named BitSeq running under R. It seems to be giving better output that some other packages.

So I downloaded the package, and gave a look to the vignette (a PDF document that can be found in the bioconductor web page containing instructions, tutorials, etc)

It turn out that BitSeq runs using SAM files. With samtools view you can easily convert your BAM files to SAM files. And I believe you don't need to extract the counts.

Now.. If you have the BAMs, have R, download BitSeq and follow the vignette instructions, you can with a little help or without that help to do your RNA-Seq analysis with the most recommended package according the consortium

[mjg]

Hi Antonio.

Have you used BitSeq? Do you know if it is possible to do DE analysis with a paired-sample design? I am considering using BitSeq and had a quick look at the vignette but I am not sure I can use this kind of design.

Thanks,
Maria

[Antonio R. Franco]

Look for details in this publication about BitSeq in Bioinformatics

In the method section you can see you can use single-end or paired-end reads

[mjg]

Thanks for the link. Sorry if it was a bit confusing, but what I meant was that I have paired/matched samples: patient 1 control and patient 1 treatment; patient 2 control and patient 2 treatment... and so on.

In edge R I can specify in the design matrix this factor, but couldn't find if or how BitSeq can incorporate this in the test.