3-5% more seq from new Roche software

[joa_ds]

I am currently using "gsRunProcessor 2.0.01.12".

For refiltering I follow the steps provided in the original manual @ page 56 and further.

Little summary of those pages:

gsRunProcessor --template=filterOnly > filterfile.xml (change the values in this xml file to your desired values)

And then to refilter:

runAnalysisFilter --pipe=filterfile.xml <your_D_folder_to_reanalyse>

[pmiguel]

I still have to install the new versions of the software, so I cannot say what the cause of the 3-5% is.

But have you checked the standard filter values compared to the older software pipeline? I know we get +50% +100% more reads when tweaking the filter settings (especially in 'exotic' experiments such as bisulphite, chip-seq or pull-down (ChIP) experiments). And most of those additional reads do map and are valid.

Nevertheless, I am quite suspicious towards improved efficiency software-based for standard experiments, especially because I got a 'hint' from a Roche tech support guy to change the value of a filter from 0.05 to 0.10 as default value, which indeed improved reads passing the filters some %s.

When it appears to be a case of filter settings, i'd recommend you to just refilter your experiments. There is a chapter on it in the manual and it takes only 10 mins or so instead of going through the whole pipeline.

Which Roche program and which parameters are you using to do this refiltering?

[joa_ds]

I still have to install the new versions of the software, so I cannot say what the cause of the 3-5% is.

But have you checked the standard filter values compared to the older software pipeline? I know we get +50% +100% more reads when tweaking the filter settings (especially in 'exotic' experiments such as bisulphite, chip-seq or pull-down (ChIP) experiments). And most of those additional reads do map and are valid.

Nevertheless, I am quite suspicious towards improved efficiency software-based for standard experiments, especially because I got a 'hint' from a Roche tech support guy to change the value of a filter from 0.05 to 0.10 as default value, which indeed improved reads passing the filters some %s.

When it appears to be a case of filter settings, i'd recommend you to just refilter your experiments. There is a chapter on it in the manual and it takes only 10 mins or so instead of going through the whole pipeline.

[pmiguel]

We 've also already seen some really interesting variants being detected. Changes from "-" to "-" for example. Not really helpful

Ahh, a zen koan variant. Sadly, those who reach enlightenment through its agency may not be disposed to post here.

--
Phillip

[Tuxido]

We haven't validated yet. We still need to check these differences. We looked up a few in the alignment browser and these seemed to look ok. We 've also already seen some really interesting variants being detected. Changes from "-" to "-" for example. Not really helpful

[pmiguel]

Yeah we also tried a few older runs this week and decided to reanalyze everything because of the improvements. We only tried DNA for now.

I think there's also some improvements in the HCDiffs detection, because we seem to pick up more differences than can be explained by the additional sequence from the improved signalprocessing and mapping

But those are real differences, not false positives? I ask, because one way to produce more "high quality" sequence is to tune a base caller to be more "confident" without actually being any more accurate.

I'm probably just being paranoid. More likely the alignment software was improved...

--
Phillip

[Tuxido]

Yeah we also tried a few older runs this week and decided to reanalyze everything because of the improvements. We only tried DNA for now.

I think there's also some improvements in the HCDiffs detection, because we seem to pick up more differences than can be explained by the additional sequence from the improved signalprocessing and mapping

[pmiguel]

We have reprocessed a few older runs using the new "Phase C" aka "v. 2.3" Roche Data Processing software that was released last week. This involves redoing image analysis/basecalling--so it is CPU-intensive. But we see more favorable results in 454BaseCallerMetrics.txt in every case we have tried. We have seen anywhere from 3% to just over 5% more totalBases.

Anyone else tried to reprocess older runs with the new software? If so, did you see improvements?

--
Phillip

PS The major improvement is that the GSAssembler (aka "Newbler") is now "cDNA-aware". EST assemblies work much better!