Advice Needed on Tuxedo Suite

[thickrick99]

Hi Everyone,

I am working on a project with two groups (A & B) each containing 5 samples. Within each sample I have two RNA-seq FASTQ reads which I downloaded. So as of now, I am planning to run tophat for each of the samples (for a total of 10 accepted_hits.bam files). From here I plan on running cufflinks for each sample with its respective .bam file to produce a total of 10 transcripts.gtf files.

My goal is to compare the differential gene expression of group A and group B. I am using the hg19 reference genome.

I have two questions. First, is there a way I can run tophat/cufflinks on all the samples all at once or do I have to run one sample, then wait until its done, then do the next sample and so on?

Second, once I have the transcripts.gtf files for each of the samples for groups A & B, is it necessary to merge them into one single .gtf file? I assume this would be done with cuffmerge right? What exactly does cuffmerge do and what are the advantages in using it. In addition, is there a need to use cuffcompare since I am comparing the differential gene expression of group A and group B or can I go straight to cuffdiff.

In general, my second question is essentially asking the pipeline for using software after tophat & cufflinks based on what I am trying to do. It would be great if someone could give me a suggestion as to the order of how I should approach this.

THANKS!

[dpryan]

If you have a cluster or single computer with sufficient RAM and cores then you can run as many instances of tophat/cufflinks simultaneously as you want.

If you don't use the -G option, then not using cuffmerge would cause you to compare apples and oranges (suppose one sample had a modified gene model and another didn't, the resulting FPKM comparison would be meaningless). If you use cuffmerge then you don't need to use cuffcompare. I think they had a methods paper with a step-by-step walk-through of an analysis, which I'd recommend you read through.

[kmcarr]

Is part of your objective to identify transcripts which have not yet been annotated in the human genome? If not then there is really no reason to use Cufflinks at all.

To perform DE analysis between groups A & B considering only already annotated genes in hg19 I would:

1. Download the hg19 reference set from the Illumina iGenomes site.

2. Prepare the known genes bowtie index as described in the TopHat manual.

3. Align your reads to hg19 with Tophat 2.x.x using the "-T" option (align to transcriptome only).

4. Count the reads aligned to genes from each of your 10 accepted_hits.bam (e.g. with feature counts or htseq-count).

5. Use the raw read counts matrix as input to your preferred DE module (e.g. EdgeR, DESeq2).