analysis of individual 454 reads?

[greigite]

We have an application for which we'd like to be able to extract information directly from 454 reads without mapping to a reference. The region we're looking at has conserved repeats interspersed with highly variable regions, and we want to extract these directly from reads and evaluate both repeat conservation and diversity of the variable regions.

Here's the question- How should we use 454 quality scores to evaluate the accuracy of variants in individual reads without mapping? Can we use them in the same way we use Sanger scores, e.g. ignoring individual bases with a cutoff score less than 20-30? My understanding of 454 quality scores, based on the 2008 Brockman paper in Genome Research, is that the algorithm used to generate them is based on validation vs a Sanger dataset. In that paper they suggest that bases with quality scores >30 have miscall errors of approx 50/Mb. We expect a higher polymorphism rate than this, so it seems like using 30 as a quality cutoff could be valid.

I'm curious to know, how do you handle analysis of base quality in individual, non-mapped reads?

[westerman]

Pretty much the way you said. Just take the 454 quality scores and use them as if they were Sanger quality scores. Unless we do not trust Roche or the Brockman paper I see no reason not to believe that the correlation is valid.

That being said, the only times I look at individual reads is after assembly is done. At the point I am sometimes interested in the reads that did not assemble. These are, by definition, the odd-ball reads to begin with and therefore getting any data from them -- on shaky evidence or not -- is a bonus for me and thus I am not overly concerned about the accuracy of the quality scores.

[greigite]

Do you know if there's been any validation of Titanium quality scores along the lines of what Brockman et al did for GS20? Is the 454 algorithm still based on this validation, or has it been updated?