Error rates in 454 FLX/Titanium reads

[Marcus]

Hi there,

I have average quality scores from several amplicon FLX and Titanium runs. Based on these postition-specific average quality scores (Q) I want to calculate postition-specific error rates/probabilities (P). If it was Sanger sequencing I could easily use the reverse Phred formula Q=-10*log(P), but I'm not sure what to use for pyrosequencing reads. Could I safely use P=10^(-Q/10)?

I read Brockman et al. (2008 Genome Research) and they say the initial quality score from GS 20 software is based on the "...probability that the base is an overcall, given the observed signal intensity for the corresponding flow". They then propose a much more comprehensive way of scoring quality, e.g. involving oberved noise in the whole read and homopolymer counts.

Does anyone know which quality scoring algorith is acutally used in FLX/Titanium these days? And does the older FLX quality scoring differ from the newer Titanium?

Many thanks in advance for any help!

Regards,
Marcus

[flxlex]

As far as I know:

- the Brockman algorithm is the current one for quality scoring (since GS FLX and software version 1.1.03)
- the scores are on the same scaling as the PHRED score, i.e. P=10^(-Q/10)
- recent versions of 454 software (gsAssembler, gsMapper) rescore 454datasets with the old scores (this can be turned off)

[Marcus]

Thanks for your answer!

So provided that the software version is not older than 1.1.03, should there be any differences at all in quality scoring between FLX and Titanium?

I have anectdotally heard (and noticed) quality differences between these two platforms, where Titanium shows worse quality. Have you noticed that as well?

Marcus

[flxlex]

Yes, same algorithm (I suppose) to determine the quality scores, but the actual quality surely is chemistry dependent. I haven't done any tests myself and am not aware of a study about this, but it would be an interesting exercise!