QC control

[Brian Bushnell]

BBMap requires a reference; all of those histograms (other than bhist) are based on mapping. But it doesn't have to be a good assembly. If you have sufficient coverage, a quick draft assembly is adequate for mapping to generate that QC data.

[Zapages]

^Is it possible to do quality control without a reference with bbmap?

I have some Ion Xpress read data. I have been told the quality control has been done, but when just checked through FastQC, it seems really bad at the end. But then again its my first time working with Ion Torrent/PGM data.

Thank you in advance for the help.

[Brian Bushnell]

I uploaded a new version of BBMap that can tell you the observed quality for claimed qualities, assuming you have a reference. You run it like this:

bbmap.sh ref=x.fasta in=reads.fastq nodisk qahist=qahist.txt

This will give 7 columns:
quality, match, sub, insertion, deletion, observed quality, observed quality (subs only)

Since deletions occur between bases, I add deletion events to bases neighboring the deletion. That would over-represent deletions, so all the other columns are multiplied by 2 to compensate. Anyway, if you plot the first column against the 6th column, it will tell you how the observed error rates correlate with quality scores.

You can get QC-relevant additional histograms with these flags:

bhist=bhist.txt (base composition by read position)
qhist=qhist.txt (average quality by read position)
mhist=mhist.txt (match, substitution, insertion, deletion events by read position)
ihist=ihist.txt (insert size distribution)

...and if you want the actual mapped reads, just add "out=mapped.sam".

[GenoMax]

My work is RNA-seq.
So the trimming step sould be 'gentle' or even not which is suggested by some experts in SeqAnswers and Biostars. Am I right?

Basically yes. Need for trimming can be evaluated based on the results you get after the first round of analysis.

[super0925]

I do not think it would be possible to equate Q-scores from Illumina/Ion by a formula because they are not using the same predictors.

Some recent studies seem to suggest that trimming may not be needed unless adapters are present or the raw qualities are poor. I would suggest doing some analysis without trimming to see what happens. BTW, What kind of analysis are you doing (re-seq, RNA-seq etc)?

My work is RNA-seq.
So the trimming step sould be 'gentle' or even not which is suggested by some experts in SeqAnswers and Biostars. Am I right?

[GenoMax]

I do not think it would be possible to equate Q-scores from Illumina/Ion by a formula because they are not using the same predictors.

Some recent studies seem to suggest that trimming may not be needed unless adapters are present or the raw qualities are poor. I would suggest doing some analysis without trimming to see what happens. BTW, What kind of analysis are you doing (re-seq, RNA-seq etc)?

[super0925]

@super0925: There is a technical note available on the ioncommunity site that describes six Quality Score predictors (and some additional ones that are apparently not listed) used for ion PGM. I did not see one specifically for proton but the one for PGM may work.

You should be able to create a free account on ioncommunity site and search for that document.

Thank you very much. I have seen that before.
The Quality Score in Proton is based on the 6 predictors which are different from Illumina. Am I right?
But I am still confused that the relationship between Proton Q score and Illumina Q score, like Q=22 Proton related to Q=30 MiSeq.
So far I could only directly judge that OK Q>22 is good reads and Q<22 is required for trimming...... Am I right?

[GenoMax]

@super0925: There is a technical note available on the ioncommunity site that describes six Quality Score predictors (and some additional ones that are apparently not listed) used for ion PGM. I did not see one specifically for proton but the one for PGM may work.

You should be able to create a free account on ioncommunity site and search for that document.

[mastal]

I think the error model for the IonProton would be more like that for 454, and involve homopolymer indels rather than substitutions.

[super0925]

These are raw quality scores, so only the manufacturers know how they are calculated. That is why platform comparisons at this level are not always the best method.

Thank you!
I agree with you. It is hard to say which machine (HiSeq/MiSeq/Proton) is better from QC score but I want to know the equivalent quality score in Proton .
Like nbahlis said
Illumina score 30 corresponding to Proton QC score 22
So how about other score e.g. 20, 40 ?

[Brian Bushnell]

Thank you!!! But I want to know the reason and why .
e.g. How do they calculate the QC score? why lower than Illumina (is that different equation?)
and, how to compare two measurement ?e,g, is that Illumina score -8 =Proton score?

If you want to determine the true quality offset, then map the reads and count the mismatch rate of bases at each quality level. I have a tool that can do that from a sam file but it's not really ready for mainstream use yet. I used it to make this plot for some recent Illumina Hiseq 2000 data, which indicates they were under-calling qualities on that run.

P.S. This only works if the error model is almost entirely substitutions. If IonTorrent reads have indels then it requires a bit more effort. I'm never used IonTorrent data.

[snetmcom]

Thank you!!! But I want to know the reason and why .
e.g. How do they calculate the QC score? why lower than Illumina (is that different equation?)
and, how to compare two measurement ?e,g, is that Illumina score -8 =Proton score?

These are raw quality scores, so only the manufacturers know how they are calculated. That is why platform comparisons at this level are not always the best method.

[super0925]

That is typical with the proton. We are told by lifetech that a22-24 score on the proton equals the 30 on the illumina platform. It has to do apparently withe the proton more conservative score calling.

Thank you!!! But I want to know the reason and why .
e.g. How do they calculate the QC score? why lower than Illumina (is that different equation?)
and, how to compare two measurement ?e,g, is that Illumina score -8 =Proton score?

[nbahlis]

That is typical with the proton. We are told by lifetech that a22-24 score on the proton equals the 30 on the illumina platform. It has to do apparently withe the proton more conservative score calling.

[super0925]

Hi all
Does proton has its own supported QC control measurement?
My reads quality looks very strange (I get this result by FASTQC software which is Illumina-based).
In Illumina we know that QC= -10*log10(p)>30 is good reads but Proton seems strange.
Does anybody know that how to measure it in Proton? Thank you!