RNA-Seq libraries ok for DGE analysis?

[JDSwenson]

Hello all,

I have spent the last several months constructing 48 libraries for RNA sequencing from developing stingray tissue, which I am hoping to sequence on a HiSeq 4000 with 100 bp PE reads.

I have been as consistent as possible throughout my library preps, but I still ended up with a range of fragment sizes after size selection (median size range from 294-485). Further, each library has a fairly broad size distribution. I've attached BioAnalyzer traces for ~ 40 of the samples.

Will these libraries be comparable enough to detect differential gene expression among them? Alternatively, are there bioinformatics tools available that can account for batch effects during processing? Fragmentation is supposed to be random, and the fragment size differences occur across treatments i.e. each treatment has replicates with both shorter and longer fragments.

Thank-you in advance for any feedback!

- John

[fanli]

I don't see anything in the traces that would make me overly alarmed. Most DGE software packages (edgeR, DESeq2, etc.) allow you to specify batch variables.

[luc]

Provided one is starting with all high quality RNA-samples, RNA-seq libraries can (and should) look indistinguishable on Bioanalyzer traces. With most kits the libraries will look similar to your sample 7A. Especially if poly-A enrichment would have been used (varying degrees of 3' biases?), we would not suggest to use these libraries -- clearly varying conditions were at work during library prep. I have no DGE data from technical replicates to back up this recommendation, though.

[huguesparri]

I advise you to build all the libraries you're going to compare at the same time to reduce technical biases.
Are you using a kit or a home made protocole?

[JDSwenson]

Many thanks for the replies and sorry it took me awhile to respond!

After consulting with a bioinformatician and the sequencing core director, we decided to sequence the samples in one lane. The tissue from which these samples came is rather rare, so even a preliminary estimate of gene expression will be valuable. If I could go back, I would have processed the samples in larger batches and used new, rather than nearly expired, AmPure beads, but there's nothing to be done at this point.

Thanks again for the feedback; fingers crossed there are some interesting discoveries to be made in these data!

[gringer]

Just watch out for barcode switching / Index swaps / cross contamination. You're going to run into it quite a bit with running 40 [presumably dual-indexed] samples on one lane.

[JDSwenson]

Originally Posted by gringer

Just watch out for barcode switching / Index swaps / cross contamination. You're going to run into it quite a bit with running 40 [presumably dual-indexed] samples on one lane.

AAHHH!! I KNOW! I just read the paper yesterday after I replied to this thread. The run finished 2 weeks ago, so there's no going back. So far, I can't think of a way to mitigate the issue. It would seem, then, that my data are likely compromised ... sigh ... It will be interesting to see how Illumina responds as the word spreads.

[gringer]

I suspect that this problem is not restricted to Illumina sequencing, and not restricted to index confusion. My suspicion is that sequences are ligating during adapter ligation steps, despite the presence of elements like A overhangs that should stop that happening.