[JueFish]

Thanks again for your thoughts, Phillip. I've been doing some comparisons between different assemblies using different approaches and it seems that using the 454 kit may be preferable to the normalization approach. With all the sequencing issues associated with it, the increase handling, and longer protocol, I'm not sure the normalization is worth it. It might save us a little cash in the long run, but the incovenience and troubleshooting might just wash all that away. From comparing, assemblies (albeit from different species/tissues), it appears the quality of data that we get from the 454 kit is much better than the normalization approach, subsequent contigs appear longer, and there's less opportunity for stuff to go wrong. You do lose something in coverage, but it's not terrible if you input with something that's ribo-depleted or poly-A selected. And, since sequence length is so important in de novo assembly, it seems trying to maximize that might be our best approach to getting a good reference assembly. Anyway, that's my current two cents, if anyone is interested.

[pmiguel]

Are you normalizing from random primed cDNA of total RNA?

I think that if you are normalizing oligo dT primed cDNA, then that would be the source of your shorter contigs, not necessarily the normalization itself.

--
Phillip

[martin2]

Anyway, we are dancing around something here and Roche is actually somewhat less mysterious than I think would normally be their wont. That is, they explicitly warn of the negative effects of polyA or polyT tracts. These tracts will blaze bright, obscuring nearby beads and possibly consuming all the dATP (or dTTP) in the flow before reaching the end of the tract. Hence the creation of the "broken oligo dT primer". This method apparently does not satisfy Roche, but their disapproval is somewhat muted.

But I think there is a second, related, issue in play. About this, Roche is more typically inscrutable, but one hears whispers. That is, one is led to believe that the image analysis software is sensitive to well-to-well bleed-over. Two beads sitting in adjacent wells of a PTP, if they emit too similar a sequence, are presumed to be "overlapping" and one or both are discarded. I'm not privy to what the threshold is for such a filter, if it does even exist, to kick in.

I think this is quite well documented in Roche manuals. I read the manuals describing signal processing tweaking possibilities and from what I remember at the very moment is that exactly for amplicon sequencing there is different processing because the sequences are synchronized between adjacent cells. Some filters are therefore turned off or increased to compensate for those previously turned off and also the width of the region from which the signal is summed up is different (wider). So, when you extract from your own machine xml templates and compare the files for shotgun vs. amplicon you may infer some knowledge and interpret it with the manual rather easily. I think some of these tricks should be probably applied to the cDNA cases when some templates contain very same sequence from their 5'- or 3'-end:


Here is a list of some differences in signal processing:

• well-screening is disabled for Ti amplicon experiments
• only EcTF group of control bead sequences is used to estimate error rates in the run instead of both groups
• most likely error subtractor has a special value for amplicons
• doValleyFilterTrimBack is disabled for amplicons
• TrimBack scale factor is stricter for amplicons than for shotgun

Here is a list of some differences in basecaller settings:

• PHRED-based quality trimming is turned off for amplicons
• flowRadius parameters is increased to 32 (do not know what is the value for shotgun)
• useCorrectionGlobalLimit is enabled for amplicon so that no n-mers can skew some algorithm calculation (maybe it is about CAFIE step?)

In summary, study the gsRunProcessor pipeline settings and play with some of the values. You can even disable dot or mixed filter, it is described in the manual as an example. Look what reads you yield with such filters disabled and then move on tuning the other parameters. I would be tempted to say that you can even to try process the cDNA runs with pre-defined amplicon settings and just see what happens.

[pmiguel]

Hi Martin2,
Yeah, I suppose were I to spend a few weeks probing the muddy depths into which Roche has sunk their gsRunProcessor pipeline I might be able to divine their intent. But thanks for your outline, I do appreciate it. This is exactly the sort of discussion I think useful. (I dare say it might happen more frequently if Roche did not feel the need to hide their manuals away from the non-devout.)
I suspect most of the issues we were having when I posted (more than 1 year ago) what you quote had to do with the 454s particular susceptibility to homopolymers. Even now the only way I know to discover if a library is "poisoned" by long polyA tails is to fire up gsRunBrowser and iterate through the early images of a run.

--
Phillip

[heleneb]

Hi everybody.
Does anyone know this kit ? http://www.tebu-bio.com/Product/035SST10096
Is seems to be very quick to prepare a cDNA library with only 50 ng of poly(A) RNA so 5µg of starting total RNA...
I am interested to know the opinion of people who have tried this kit...
Many thanks !
Helene