Broad's Experience with the MiSeq...300bp reads possible!

[ECO]

Just got this in email, a webinar from the Broad describing their initial year of work with the MiSeq.

http://www.avitage.com/proc/collections/w/95a6c1ce5860410085a7daa914d7f7e9/index.html

The highlight for me is they present data from a 300bp single-end run! See below for a screenshot of the slide.

[MrGuy]

Maybe I'm being stupid here, but an error rate of 0.4% and a Q30 score are not congruent. I'm used to the Q scores of sanger/phred.

http://en.wikipedia.org/wiki/Phred_quality_score

They had some slides in there stating that they are getting Q52. This means they are getting greater than 99.999% accuracy IFF they are following PHRED (ie, 0.001% error at 1x coverage which would be great if this is true). However, just looking at a >Q30 posted above, the error rate is 0.4%. This means the real Q score is closer to Q24 if they error rate is true.

What am I missing. Can anyone explain?

[Vinz]

Thanks for the link to this great presentation. Data looks impressive.

@MrGuy
If you have 100% Q30 values you should have a error rate of 0.1%. However, they state >65% Q30 bases. That leaves up to 35% with lower Q-values. Assuming 10% of bases at Q10, 90 % at Q30 you would expect 0.1 * 10% + 0.9*0.1%=1.09% error rate.

[ECO]

Running FastQC on my fresh 300bp run now...

[GenoMax]

Eco: Please post info about mapability of the sequences once you complete the analysis (if this is a re-sequencing run).

Running FastQC on my fresh 300bp run now...

[ECO]

I'll do what I can...here is the QV distribution and raw/PF yield numbers.

[pmiguel]

Hi Eco,
Interesting! How long did the run take? Any special requirements for doing 300 cycles? Or did you just add extra reagents and go?

I also notice that you have very high %PF at a cluster density of nearly 1000K/mm2. Is that what you usually see?

--
Phillip

[NextGenSeq]

To me it looks like the last 100 cycles are a waste of reagents.

[MrGuy]

Thanks for the link to this great presentation. Data looks impressive.

@MrGuy
If you have 100% Q30 values you should have a error rate of 0.1%. However, they state >65% Q30 bases. That leaves up to 35% with lower Q-values. Assuming 10% of bases at Q10, 90 % at Q30 you would expect 0.1 * 10% + 0.9*0.1%=1.09% error rate.

Ah, got it. Do you know what look-up tables are they using for their Q values? How does this really compare to "old gen" sequencing? Is there some kind of normalization factor you can multiply with? I'm just trying to get a handle on all of this and don't quite understand how Q values of all these different platforms really compare.

Thanks! I do like the long reads...

[ECO]

Hi Eco,
Interesting! How long did the run take? Any special requirements for doing 300 cycles? Or did you just add extra reagents and go?

I also notice that you have very high %PF at a cluster density of nearly 1000K/mm2. Is that what you usually see?

Not even addition of extra reagents...the 300 cycle kit was fine. Our instruments still throw an error when running 2x151 + 6 cycle index reads (not enough reagents...) which is a bug that will be fixed soon.

The PF% was typical...here are a few more typical runs.

To me it looks like the last 100 cycles are a waste of reagents.

Last 80...

[LVAndrews]

Ran 300bp read on v2 chemistry

We received our MiSeq in October and when Illumina came to perform the "training run" with PhiX control, I told them we would be running 300bp in one direction to see how the quality may have improved over the v1 data from ECO and Broad. We had to put in a few reads the other direction, so the instrument was configured for 301 reads the first direction, then 6 from the other. Here are some screenshots:








Had a little problem getting the run summary up here. Maybe this is enough for now?

Andy

[yaximik]

What average library size was in Broad run? PhiX control has a quite narrow distribution around 500 if I remember that correctly, so in this case 300 cycles descend below fragment middle, yet still far enough from the floor. Looks like Broad's sample was shorter, was it? I also wonder how thumbnails looked at cycles 30 and 300, considering drop in intensity there were substantially fewer active clusters