[thinkRNA]

Hi Jay, thanks for getting back to me so quickly. I have certainly thought about PCR duplicates being the source of variation as we did PCR amplify before sequencing. The problem is my data is single end Illumina data and I don't know how to differentiate whether a duplicate read is a result of PCR amplification or a genuine indication of a copy of the mRNA. I think if I had paired end reads, I could use the size distribution of the library to eliminate PCR duplicates. I can perhaps apply the assumption of the library size to single end reads too, but I need to think more about this? If any one has other ideas on how to detect PCR duplicates in RNA seq data, please let me know.

Also, I am probably missing something but since you didn't find anything with free energy, how are you convinced that RNA sec structure interfering with the transcriptase could be a source? Are you thinking the algorithm for detecting free energy is not efficient? if the RNA is sheared before creating the cDNA, I think it should eliminate sec structure formations (though I could be wrong).
Finally, this is a stupid question, at what step is the GC coverage variation introduced?

[thinkRNA]

I guess you don't need to answer the last stupid question as I found a paper that explains it really well.

http://nar.oxfordjournals.org/cgi/content/full/36/16/e105

[jay]

I didn't get anywhere with free energy, possibly because of the sequence we are using - a 10kb RNA genome - I found tools which would predict its shape, and that would predict free energy for shorter sequences, but I couldn't find a tool quickly online that would give me a free energy estimate per base for such a long sequence, so I gave up after a day or so of looking, as we were not interested in quantification per se. If you have any ideas of a good tool for this I'd definitely be interested to give it a go. You may be right about shearing controlling for sec structures, I will talk to our experimentalist about this, as his thought was that secondary structure would be an issue.

[thinkRNA]

have you tried mfold?

http://mfold.bioinfo.rpi.edu/

I haven't used it but a colleague says its the one of the best out there.
EDIT: it takes max 1500 bases, so I guess you'll have to chop you RNA sequence (!)

[Xi Wang]

There are two papers studying on RNA-seq biases:
Modeling non-uniformity in short-read rates in RNA-Seq data. Genome Biology.
Biases in Illumina transcriptome sequencing caused by random hexamer priming. NAR.

They may help you on this topic.