View Single Post
Old 04-25-2012, 04:36 AM   #15
Senior Member
Location: Birmingham, UK

Join Date: Jul 2009
Posts: 356

Originally Posted by pmiguel View Post
Still, in case you do another one, my (unsolicited) input: I personally would like to see a dollar-for-dollar constrained test and an hour-for-hour constrained test of these instrument systems.
Yep, I think that's something we'd like to do in an ideal world. I think more likely we can improve on this test by turning it into more of a community effort - and not just for E. coli but for more complex genomes, high/low GC genomes, amplicons, exomes etc which is obviously important to people.

I'll post some ideas up when I've had time to formulate the idea a bit more.

Originally Posted by pmiguel View Post
If I could switch the topic somewhat. In figure S2, in this pdf, it looks like the Torrent and the MiSeq both slightly undervalue the quality of their base calls for most quality values and the 454 Jr slightly overvalues the quality of their base calls. Am I reading the chart correctly? I knew the Torrent tended towards "modesty" in these matters. But I had the vague impression that Illumina tended towards "bluster".
Yes, you are reading that chart correctly and your suspicions are generally correct.

That plays out when you calculate measured quality for Ion Torrent PGM (actual quality better than expected) and 454 Jr (actual quality lower than expected) (Figure 1). I guess what was surprising to me when I did this was that the PGM quality caller is actually better than many had thought - it had been reported on blogs that it was seriously underestimating accuracy, but turns out that was because only nucleotide substitutions were counted in the recalculation process (something to watch out for with GATK).

For MiSeq the actual quality is a bit down on expected. But bearing in mind Phred's log-scale it turns out its hardly noticeable when plotted on a linear scale of % accuracy, whereas the effect is much more obvious for the other two platforms. So I guess I am less trustful of the top end of the quality values being precisely correct from MiSeq.
nickloman is offline   Reply With Quote