workflow question

[spacup]

Hi all,

I am a newbie in RNA-seq and I am wondering whether the workflow I imagined is correct or not :
As Mouse lemur genome is not well documented and reads aligned poorly, I thought I could map unmapped reads on macaca genome then next unmapped reads on chimpanzee genome and then next unmmaped reads on human genome, to have the maximum reads mapped. Finally, I would merge all the sam file obtained to make DE analysis.

Does it sound good? Can I merge different Sam files obtained from mappings in different animal species in a unique file? I guess I can because as I always map unmapped reads from previous mappings, I will no have redundancy. The problem is that genes ID will be from different species and I don't know if it's possible to analyse that in a DE analysis...

thanks.

[swbarnes2]

In general, it would be better for you to throw the unmapped reads at all of those genomes at once, rather than one at a time, if possible. (I don't know the max sizes for bwa or bowtie, and it might depend on your computer specs too.)

If a read maps to the macaca genome, but maps better to the human genome, wouldn't you rather see it reported as a human genome hit?

The other thing you can try is de novo transcriptome assembly on the first set of unmapped reads.

And yes, if you do it separately, you can merge all those .bam files together. Concatenate all the headers together, then use that with samtools merge to merge the .bams

[Wallysb01]

If you have decent read depth, I would suggest using Trinity's de novo assembly and downstream analysis pipeline over all this. It sounds like a huge mess that will be very difficult to unravel once you have these alignments.

If you wanted you could take this a step further and do another thing, which would be to do a reference guided transcriptome assembly, then merge with your de novo assembly using CD-HIT to remove redundancy. Then map the reads to this transcriptome using the trinity post-assembly analysis packages.

That way you might be able to get better expression data on lowly expressed genes that won't de novo assemble well.

[spacup]

Thank you for these answers.
The problem with de novo assembly is that my computer is not enough powerfull for this, I think....

What do you mean by read depth and how do you calculate it ?
Is it : number of reads * read length / transcriptome size ?

[Wallysb01]

Read depth for transcriptomes is a bit funny, because it depends on expressed genes. But I'd say you're doing pretty good if you have 50M reads per sample, including replicates. It helps if your RNA-seq libraries are strand specific and ribo-depleted or poly-A selected, but isn't necessary.

As for your computer, you should be able to find clusters were de novo transcriptome assembly is possible. It isn't as bad as de novo genome. Any computer/node with 128GB of RAM, which is getting pretty common, will usually do. Check out DIAG for example. http://diagcomputing.org

[spacup]

Actually I have about 40M reads by samples and the library is poly-A selected. Furthermore, I have 2 or 3 biological replicates by condition.

I will take a look to this link.
Thanks a lot for your help.