Correlation between library construction protocols

[Asaf]

Hi,
We've done some experiments in out lab and from the same RNA constructed libraries using two different protocols. To our surprise the correlation between the two protocols was lower than correlation of biological replicates (0.9 vs 0.93, spearman r). The correlation between the log-fold change of genes between two conditions using the two prep protocols was extremely low (about 0.3).
Have anyone tried to compare correlation of different library preparation protocols? Is something like that was published? And, how can I decide which protocol to use, the one that gave better correlation between replicates or is there a better way to assess this?
Thanks

[NextGenSeq]

We've found that the most critical step is the cDNA synthesis.

We compared Illumina and Ion Torrent RNA-Seq data and have gotten a correlation coefficient of over 0.99 if we use the same cDNA to make the libraries.

[snf]

I was just talking with our facility manager yesterday about this, who claims to have done similar analysis and found similar results (poor correlation between libraries prepared with different protocols). He indicated the following things are critical to compare between library preps:

-Fragmentation in RNA or cDNA
-amplification method
-blunt or sticky end ligation (adapter)

This makes a lot of sense to me considering hidden specificity may be present in many of these steps. That said, it is disconcerting when comparing published datasets to each other or to current data. Worth keeping in mind for sure. I was wondering if this could be added as a factor to a generalized linear model in software like DEXSeq to possibly ameliorate the effect but haven't tested this yet.

[Asaf]

Thanks, it's good to hear we're not alone. A lot of steps were different between the two protocols including cDNA synthesis which explains the differences. I'll just add another factor that we learned that should be taken into consideration - size selection, different protocols select for different sizes.

[snf]

You might try comparing GLM's with or without a term for the library prep method to see if you can correct for this post-hoc, if it's helpful.

[NextGenSeq]

I would also be suspicious of the ligation.

miRNA-Seq is notorious for showing ligation bias. Some sequence motifs are virtually impossible to ligate.